1

Transmitter and Receiver Architectures for Molecular Communications: A Survey on Physical Design with Modulation, Coding and Detection Techniques Murat Kuscu, Student Member, IEEE, Ergin Dinc, Member, IEEE, Bilgesu A. Bilgin, Member, IEEE, Hamideh Ramezani, Student Member, IEEE, Ozgur B. Akan, Fellow, IEEE,

Abstract—Inspired by Nature, molecular communications infrastructure [1], [2]. In that respect, MC is promising for (MC), i.e., use of molecules to encode, transmit and receive infor- novel applications especially towards information and commu- mation, stands as the most promising communication paradigm nication technology (ICT)-based early diagnosis and treatment to realize nanonetworks. Even though there has been exten- sive theoretical research towards nanoscale MC, there are no of diseases, such as continuous health monitoring, smart drug examples of implemented nanoscale MC networks. The main delivery, artificial organs, and lab-on-a-chips [3], [4] (see reason for this lies in the peculiarities of nanoscale physics, Fig. 1(a)). It bears a significant potential as an alternative to challenges in nanoscale fabrication and highly stochastic nature conventional wireless communication in environments where of biochemical domain of envisioned nanonetwork applications. the latter may fail, e.g., intrabody medium [5] and confined This mandates developing novel device architectures and com- munication methods compatible with MC constraints. To that channels like pipe networks [6], [7]. end, various transmitter and receiver designs for MC have Discrete nature of information carrying agents, i.e., been proposed in literature together with numerable modulation, molecules, and peculiarities arising from the nanophysical and coding and detection techniques. However, these works fall into biochemical processes, and computational and energy-based domains of a very wide spectrum of disciplines, including but limitations of communicating nanomachines give rise to novel not limited to information and communication theory, quan- tum physics, materials science, nanofabrication, physiology and challenges that cannot be always tackled by the conventional synthetic biology. Therefore, we believe it is imperative for the ICT methods, which are mostly tailored for electromagnetic progress of the field that, an organized exposition of cumulative means of communications with macroscale components. This knowledge on subject matter be compiled. Thus, to fill this gap, requires rethinking of the existing ICT tools and devising in this comprehensive survey we review the existing literature new ones for MC in the light of envisioned applications on transmitter and receiver architectures towards realizing MC amongst nanomaterial-based nanomachines and/or biological and considering especially available material and fabrication entities, and provide a complete overview of modulation, coding technologies, which basically set the limitations. and detection techniques employed for MC. Moreover, we identify MC has been extensively studied from various aspects the most significant shortcomings and challenges in all these over the last decade. The research efforts are mainly cen- research areas, and propose potential solutions to overcome some tered around developing information theoretical models of of them. MC channels [8], [9], devising modulation and detection Index Terms—Molecular communications, Nanonetworks, In- techniques [10], [11], and system-theoretical modeling of MC ternet of Nano Things, Transmitter, Receiver, Modulation, Cod- applications [12], [13]. For the physical design of MC trans- ing, Detection mitter (Tx) and receiver (Rx), mainly two approaches have arXiv:1901.05546v1 [cs.ET] 16 Jan 2019 been envisioned: biological architectures based on engineered I.INTRODUCTION bacteria enabled by synthetic biology, and nanomaterial-based OLECULAR communications (MC) is a bio-inspired architectures, as shown in Fig. 1(b) [1]. However, none of M communication method that uses molecules for en- these approaches could be realized yet, and thus, there is no coding, transmitting and receiving information, in the same implementation of any artificial micro/nanoscale MC system way that living cells communicate [1]. Since it is inherently to date. As a result, the overall MC literature mostly relies bio-compatible, energy-efficient and robust in physiological on assumptions isolating the MC channel from the physical conditions, it has emerged as the most promising method to processes regarding the transceiving operations, leading to a realize nanonetworks and Internet of Nano Things (IoNT), i.e., plethora of ICT techniques, feasibility and performance of artificial networks of nanoscale functional units, such as nano- which could not be validated. biosensors and engineered bacteria, integrated with the Internet Our objective in this paper is to tackle this discrepancy between the theory and practice by providing a comprehensive The authors are with the Internet of Everything (IoE) Group, Electrical account of the recent proposals for the physical design of Engineering Division, Department of Engineering, University of Cambridge, MC-Tx/Rx and the state-of-the-art in the theoretical MC re- UK, CB3 0FA (e-mail: {mk959, ed502, bab46, hr404, oba21}@cam.ac.uk). This work was supported in part by the ERC projects MINERVA (ERC- search covering modulation, coding and detection techniques. 2013-CoG #616922), and MINERGRACE (ERC-2017-PoC #780645). We provide an overview of the opportunities and challenges 2

(a) (b) Engineered Bacteria-based Biological MC-Tx/Rx MC-based Alzheimer & Epilepsy Monitoring and Brain Stilmulation Networks

MC-based Information Body-Area Nanosensor Molecules Networks Transmitter MC Channel Receiver Heart Transmitted Received Monitoring Signal Signal INTERNET MC Networks Information Molecules Cancer Monitoring/ Drug Delivery MC Networks

Healthcare Nano/Macroscale Provider (or Bio-Cyber) Interface Electrical Stimuli-responsive Graphene Biosensor-based Hydrogel/Graphene MC-Tx MC-Rx

Fig. 1: (a) An intrabody continuous healthcare application of IoNT enabled by MC nanonetworks [1]. (b) Components of an MC system with biological and nanomaterial-based MC-Tx/Rx design approaches, which are reviewed in Section II and Section III. regarding the implementation of Tx/Rx and corresponding MC receiver. It is shown that field effect transistor (FET)- ICT techniques which are to be built on these devices. based biosensors (bioFETs) satisfy the requirements of an MC Although we mostly focus on diffusion-based MC in this receiver, i.e., nanoscale dimension, in-situ, label free, contin- review, we also partly cover other MC configuration, such as uous and selective detection of information carriers. Thus, in MC with flow, microfluidic MC, bacteria conjugation-based this paper, we mostly focus on the operation principles of MC, and molecular-motor powered MC. We first investigate bioFETs and their main design parameters that must be se- the fundamental requirements for the physical design of mi- lected to optimize performance of the device as an MC-Rx. We cro/nanoscale MC-Tx and MC-Rx, e.g., those regarding the review the possible types of bioreceptors, e.g., natural receptor energy and molecule consumption, computational complexity proteins, aptamer/DNAs, which act as the interface between and operating conditions. In light of these requirements, both the MC channel and the receiver. We also study available of the design approaches previously proposed for MC-Tx/Rx, options for the nanomaterials, e.g., silicon nanowire (SiNW) i.e., artificial Tx/Rx designs enabled by the discovery of new and graphene, used as the transducer channel, and evaluate nanomaterials, e.g., graphene, and biological Tx/Rx architec- their feasibility and potential to be a part of MC receiver tures based on the engineering tools provided by synthetic in terms of mechanical, electrical and chemical properties biology, are covered. and associated fabrication challenges. We then provide an overview of studies focusing on the modeling and performance The literature on physical design of micro/nanoscale artifi- evaluation of bioFETs from MC perspective. cial MC-Tx is almost nonexistent, except for a few approaches focusing on standalone drug delivery systems [14] and mi- Regarding biological architectures, there are more opportu- crofluidic neural interfaces with chemical stimulation capabil- nities enabled by the synthetic biology tools in engineering ities [15], [16], which, however, do not address the commu- bacteria with artificial MC-Tx/Rx functionalities. Biological nication theoretical challenges of MC. Therefore, regarding device architectures, in general, have the advantage of bio- the MC-Tx architecture, we mostly investigate and propose compatibility, a necessary condition for many MC applications several potential research directions towards nanomaterial- within healthcare and Internet of Bio-Nano Things (IoBNT) based implementation of MC-Tx based on recent advance- paradigm [1], as well as availability of already existing molec- ments in nanotechnology, novel materials and microfluidics. ular machinery for sensing, transmitting and propagation at We will particularly focus on the utilization of nanoporous micro/nanoscale, all of which are prone to be utilized for MC graphene membranes in microfluidic settings to control the nanonetworking. In terms of Tx architectures, biological-based release of information molecules through electric field. For approaches also prove advantageous thanks to the capability the design of artificial MC-Rx, we will focus on nanomaterial- of harnessing transmitter molecules or synthesizing them from based biosensing approaches, as a biosensor and an MC-Rx readily available molecules, which resolves the reservoir prob- have similar functionality, i.e., both of them aim to detect lem. Biological architectures considered in this review include the physical characteristics of molecules in an environment. the use of genetically engineered flagellated bacteria, e.g., Previously in [17], the structure and performance of available E. coli and viruses as encapsulated carriers of information, biosensing approaches, i.e., electrical, optical and chemical e.g., RNA or DNA, bacteria with engineered gene circuits sensing, are studied to evaluate their feasibility as an artificial for protein-based communication and Ca++ circuits for hor- 3 monal communication, as well as other mechanisms such (BERs). Novel codes that specifically handle with ISI need as microtubule networks and biological-nanomaterial hybrid to be developed for mitigation of these errors. Moreover, approaches. For biological receivers, synthetic biology tools in general, energy is a scarce resource for communicating allow performing complex digital and analog computations for agents in a nanonetwork, requiring implementation of low detection [18], integrate computation and memory [19] and complexity codes with low computational cost, which renders enable continuously keep track of individual receptor states, the application of state-of-the-art codes, such as Turbo codes as naturally done by living cells. In this paper, we provide a [23], unfeasible. Consequently, many researchers have focused review of available studies on the use of genetic circuits for on simpler block codes, such as Hamming codes [24], whereas implementing MC receiver architectures. The major challenge some have devised novel channel codes specifically for MC. In in use of genetic circuits for implementing MC-Tx/Rx func- this review, we give a comprehensive account of these channel tionalities in engineered bacteria arises from the fact that in- codes. formation transmission in biological cells is through molecules Detection is by far the most studied aspect of MC in the and biochemical reactions. This results in nonlinear input- literature, but there are still many open issues mainly because output behaviors with system-evolution-dependent stochastic of the unavailability of any MC-Rx implementation that can effects that are needed to be comprehensively studied to validate the developed methods. This lack of physical models evaluate the performance of the device. This paper will present for realistic devices has led the researchers to make simplifying a brief account of these challenges and opportunities regarding assumptions about the sampling process, receiver geometry the use of biological architectures. and channel and reception noise. Depending on the type of We also provide an overview of the communication tech- these assumptions, we investigate MC detection methods under niques proposed for MC. In particular, we will cover modu- two categories: detection with passive/absorbing receivers and lation, coding and detection methods. Modulation techniques detection with reactive receivers. We provide a qualitative in MC fundamentally differ from that in conventional EM comparison of these methods in terms of considered chan- communications, as the modulated entities, i.e., molecules, nel and receiver characteristics, complexity, type of required are discrete in nature and the developed techniques should be channel state information (CSI), and performance. robust against highly time-varying characteristics of the MC In summary, this paper provides design guidelines for channel, as well as inherently slow nature of the propaga- the physical implementation of MC-Tx/Rxs, an overview of tion mechanisms. Exploiting the observable characteristics of the state-of-the-art MC methods on transceiving molecular molecules, researchers have proposed to encode information messages, and a detailed account of the challenges and fu- into the concentration, e.g., concentration shift keying (CSK) ture research directions. We believe that this comprehensive [20], molecule type, e.g., molecule shift keying (MoSK) [10], review will help researchers to overcome the major bottleneck and molecule release time, e.g., release time shift keying resulting from the long-standing discrepancy between theory (RTSK) [21]. However, the proposed modulation schemes and practice in MC, which has, so far, severely impeded the generally suffer from low data rates as it is not practical to use innovation in this field linked with huge societal and economic modulation schemes with high number of symbols due to inter- impact. symbol interference (ISI), detection sensitivity and selectivity The remainder of this paper is organized as follows. In problems. Therefore, in this paper, we also investigate two new Section II, we investigate the design requirements of an MC- research directions that can potentially alleviate these prob- Tx, and review the physical design options for nanomaterial- lems: (1) utilization of DNA/RNA as carrier molecules such based and biological MC-Tx architectures with an overview that nucleotide strands differing in physical properties, e.g., of state-of-the-art MC modulation and coding techniques. We length, dumbbell hairpins, short sequence motifs/labels and focus on the opportunities for the physical design of an MC-Rx orientation, can enable transmitting high number of symbols in Section II, where we also provide a comprehensive review that can be selectively received, (2) encoding large amount of of the existing MC detection schemes for passive, absorbing information into the base sequences of DNAs, i.e., Nucleotide and reactive receivers. In Section IV, we outline the challenges Shift-Keying (NSK) [22], to boost the data rate up to the order and future research directions towards the implementation of of Mbps such that MC can compete with traditional wireless MC Tx/Rxs, and the development of the corresponding coding, communication systems. modulation, and detection techniques. Finally, we conclude the Another fundamental aspect of MC is coding. In a commu- paper in Section V. nication system coding is done at two stages, source coding and channel coding. Source coding is related to the structure II.MOLECULAR COMMUNICATION TRANSMITTER of information to be transmitted, which has so far not been discussed in MC literature, and is not covered in this review. MC transmitter (MC-Tx) encodes information in a physical On the other hand, various channel coding schemes have been property of molecules such as concentration, type, ratio, order applied to MC channels with the hope of error correction. or release time, and releases information molecules (IMs) However, conventional channel coding schemes developed for accordingly. To this aim, information to be transmitted is classical EM-wave based communications are incompatible required to be mapped to a sequence of bits through source and with nanoscale MC in many aspects. To start with, channel channel coding to represent the information with less number characteristics are very different, where diffusive channels of bits, and to introduce additional bits to the information suffer extremely from ISI, which causes high bit error rates with the purpose of providing error correction, respectively. 4

Then, the modulator unit encodes the information in a property [37], the issue of biocompatibility of these devices stands out of molecules and control the release of IMs according to a as one of the most challenging research issues to be addressed. predetermined modulation scheme. Finally, a power source Moreover, corrosion of implants inside body is another com- and IM generator/container are required to provide energy monly known issue [38] restricting device durability, which and IM molecules for MC-Tx. The interconnection of these due to increased surface to volume ratio at nanoscale has an components in an MC-Tx is illustrated in Fig. 2. In this effectively amplified effect on nanoscale implants compared to section, we first discuss the requirements of an MC-Tx and their larger versions. In addition to biochemical toxicity and investigate the utilization of different IMs. Then, we review durability, flexibility of implanted devices is another important the available approaches in physical design of MC-Txs, which aspect of biocompatibility, as mechanical mismatch between can be categorized into two main groups, (i) nanomaterial- soft biological tissue and the implanted device triggers in- based artificial MC-Tx and (ii) biological MC-Tx based on flammation followed by scar tissue formation [39], which synthetic biology. eventually restricts the implant to carry out its intended task. Accordingly, the last decade has witnessed extensive research on novel biocompatible materials such as polymers [38], soft A. Design Requirements for MC-Tx organic electronics [40], dendrimers, and hydrogels [41]. It is 1) Miniaturization: Many novel applications promised by imperative to note that, all issues of biocompatibility men- MC impose size restrictions on their enabling devices, re- tioned here specifically apply to nanomaterial-based device quiring them to be micro/nanoscale. Despite the avalanching architectures, and they can be almost completely avoided progress in nanofabrication of bioelectronics devices over by the choice of a biological architecture, e.g., genetically the last few decades [25], fabrication of fully functional engineered organisms [1]. nanomachines capable of networking with each other and their 4) Transmission Performance: Performance of a molecu- surroundings in order to accomplish a desired task still evades lar transmitter should also be evaluated by how much control us [26]. Moreover, with miniaturization the surface area to it has over the transmission of molecules. Important per- volume ratio increases, causing surface charges to become formance metrics include off-state leakage, transmission rate dominant in molecular interactions. This causes the behavior precision, resolution and range, as well as transmission delay. of molecules passing through openings with one characteristic Off-state leakage, or unwanted leakage of molecules from dimension less than 100nm to be significantly different than transmitter, from MC perspective, degrades communication that observed in larger dimensions [27]. Consequently, as the significantly by contributing to background channel noise. On release aperture of many considered molecular transmitter the other hand, it is unacceptable for some niche applications, architectures are commonly in nanoscale, peculiarities arising such as targeted drug delivery [42], where leakage of drug from this phenomenon need to be taken into account in molecules would cause undesired toxicity to the body. During transmitter design. operation, control over rate of transmission is essential. For 2) Molecule Reservoir: The size restriction introduces, instance, in neural communications information is encoded, besides the need for energy harvesting due to infeasibility of among other fashions, in the number of neurotransmitters deploying a battery unit, which is a problem for the whole released at a chemical synapse, and accordingly transmitter nanomachine in general, another very important problem for architectures envisioned to stimulate neurons via release of any transmitter architecture, namely, limited resources of IMs neurotransmitters, e.g., glutamate [43], will need to encode [28]. Typically, a transmitter module would contain reservoirs, information in neurotransmitter concentrations via very precise where the IMs are stored to be released. However, at dimen- transmissions. In this respect, transmission rate precision, sions in question, without any replenishment these reservoirs resolution and range are all variables that are determinant will inevitably deplete, rendering the nanomachine function- in the capacity of communication with the recipient neuron. ally useless in terms of MC. Moreover, the replenishment rate Finally, transmission delay is in general a quantity to be of transmitter reservoirs has direct effect on achievable com- minimized, as it contributes to degradation in communication munication rates [29], [30]. Possible theoretically proposed rates. Moreover, for instance in neural communications, where scenarios for reservoir replenishment include local synthesis of information is also encoded in frequency of signals, there is an IMs, e.g., use of genetically engineered bacteria whose genes upper bound on permissible transmission delay of a transmitter are regulated to produce desired transmitter proteins [31], as in order to satisfy a given lower bound on maximum frequency well as, employment of transmitter harvesting methods [32]. transmittable. Corresponding technological breakthroughs necessary for re- alization of these approaches are emerging with advancements in the relevant fields of genetic engineering [33], and materials B. Physical Design of MC-Tx engineering [34], [35]. MC uses molecules for information transfer unlike the tra- 3) Biocompatibility: Most profound applications of nan- ditional communication systems that rely on electromagnetic odevices with MC networking capabilities, such as, neural (EM) and acoustic waves. Therefore, the physical design of prosthetics, tissue engineering, targeted drug delivery, BMIs, MC-Tx significantly differs from the traditional transmitters. immune system enhancement, etc., require the implantation The key components of an MC-Tx is illustrated in Figure of corresponding enabling devices into biological tissue [36]. 2. The first element in MC-Tx is the information source Combined with increased toxicity of materials at nanoscale that represents either a bit-wise information generated by a 5

channel. There are three main propagation channels that can Power Source IM Generator/ (or EH) Container be exploited by MC: free diffusion, flow-assisted propaga- tion, and motor-powered propagation [6], [47]. Diffusion is the movement of molecules from high concentration to low Processing Unit concentration with random steps. This process requires no Information Source/Channel Modulation IM Release external energy and the molecules travel with the thermal Source Coding energy of the medium, e.g., liquid or air. Diffusion process can be modeled with the Fick’s law, i.e., a partial differential equation calculating the changes in molecular concentration in Fig. 2: Physical architecture of an MC transmitter. a medium, as ∂c = D∇2c, (1) ∂t nanomachine or a biological signal from living cells such as where c is the molecule concentration, and D is the diffusion neurons and cardiomyocytes. coefficient, which depends on the temperature, pressure and The next building block in an MC-Tx is the processing unit, viscosity of the medium. Calcium signaling between cells which performs coding, modulation and control of IM release. and propagation of neurotransmitters in synaptic cleft are This block may not appear in biological systems utilizing good examples of biological systems utilizing diffusion for MC in a basic way such as hormonal communication inside molecular information transfer. Diffusion-based MC provides human body, where the information is transferred via a single short range information transfer in the order of micrometers molecule type with on-off keying. For this purpose, MC-Tx [6]. processing unit can operate via chemical pathways [6] or Second type of MC channel is flow-assisted propagation, in through biocompatible micro-scale processors, or via synthetic which molecules propagate via both diffusion and flow, such genetic circuits [44]. In case of biological Tx architectures, as flow in the blood stream and air flow created by wind. processing units performing logic gates and memory functions Flow-assisted propagation can be modeled with the advection- in the form of synthetic genetic circuits can be embedded into diffusion equation, which is represented as cells [44], [45]. ∂c Inside the processing unit, the first step is to represent + ∇.(vc) = D∇2c, (2) analog or digital information with the least number of bits ∂t possible through source coding. Then, channel coding block where v is the velocity vector for the flow in the medium. introduces extra bits to make data transmission more robust The diffusion case, flow-assisted propagation does not require against error-prone MC channels. After the coding step, digital any external energy source, and uses the thermal energy information is transformed into an analog molecular signal that exists in the propagation medium. Communication range according to a predetermined modulation scheme. There are of flow-assisted propagation can reach up to several meters various modulation schemes proposed for MC by using molec- as in hormonal communications in the blood stream and ular concentration, type of molecules or time of molecule communications via pheromone between plants [6]. release. The detailed discussion of modulation techniques Lastly, motor-powered propagation includes molecular mo- for MC can be found in Section II-C. The processing unit tors and bacteria with flagella, e.g., E. coli. Molecular motor- controls IM release according to the data being transmitted powered microtubules, also known as walkway-based archi- and the modulation scheme. IM molecules are provided from tectures [48], [49], use predefined microtubules with motor IM generator/container, which can be realized as either a proteins, which carry IMs from one end to other end of the reservoir containing genetically engineered bacteria to produce tube. Motor proteins such as Kinesis consume external energy IM molecules on demand, or a container with limited supply in the form of adenosine triphosphate (ATP) to move IMs in of IM molecules. microtubules with the diameters of 20-30nm [50], [51]. This There are two types of molecule release mechanisms [46]: motion can be mathematically represented as [6] instant release, where certain number of molecules (in mol) x = x + v ∆t, (3) are released to the medium at the same time, and continuous i i−1 avg release, where the molecules are released with a constant where x is the location of the molecule inside the microtubule, rate over a certain period of time (mol/s). Instant release can vavg is the average speed of molecules moving via molecular be modeled as an impulse symbol, while continuous release motors, and ∆t is the time step. Molecular motors enable can be considered as a pulse wave. According to applica- medium range molecular communication up to a millimeter tion requirements and channel conditions, one of the release [48]. In addition, [47], [52] propose a nanonetwork archi- mechanisms can be more favorable. Bit-wise communications, tecture in which engineered bacteria with flagella is utilized for example, require rapid fade-out of IMs from the channel to carry IMs for medium range MC (µm-mm). As in the as these molecules will cause ISI for the next symbol, thus, molecular motor case, bacteria consume energy for movement instant release is more promising. In addition, alarm signals for via flagella. warning plants and other insects can favor continuous release The processing unit in MC-Tx may require a power source to increase detection probability and reliability. for performing coding and modulation operations especially MC-Tx architecture also shows dependency on propagation in the nanomaterial-based MC-Txs. Battery-powered devices 6

Transmitter Channel Receiver A

IOpen

i Current Information I Molecule Blocked Time Fig. 3: System model for the DNA-based MC. suffer from limited-lifetime and random battery depletions, the existence of hydroxyl groups in RNA makes it less stable i.e., unexpected exhaustion of sensor battery due to hardware to hydrolysis compared to DNA. Hence, DNA can be expected or software failure, that significantly affect the reliability of to outperform RNA as an information carrying molecule. systems that rely on MC. Therefore, MC-Tx unit is required to Recent advancements in DNA/RNA sequencing and syn- operate as a self-sustaining battery-less device by harvesting thesis techniques have enabled DNA-encoded MC [61]–[63]. energy from its surrounding. For this purpose, energy har- For information transmission, communication symbols can be vesting (EH) from various sources, such as solar, mechanical, realized with DNA/RNA strands having different properties, chemical, can be utilized in MC-Tx architectures. Hybrid i.e., length [64], dumbbell hairpins [61], [65], short sequence energy harvesting techniques can be employed as well to motifs/labels [62] and orientation [66]. For information de- increase the output power reliability [53]. Concerning the tection, solid-state [62], [67] and DNA-origami [68] based intrabody and body area applications, human body stands as nanopores can be utilized to distinguish information symbols a vast source of energy [54], which has been exploited to based on the properties of DNA/RNA strands by examining the power biomedical devices and implants in many ways, e.g., ionic current characteristics during translocation, i.e., while thermoelectric EH from body heat [55], vibrational EH from DNA/RNA strands pass through the nanopores as illustrated heartbeats [56], and biochemical EH from perspiration [57]. in Fig. 3. The utilization of nanopores for DNA symbol Nevertheless, design and implementation of EH methods at detection also enables the miniaturization of MC-capable micro/nanoscales for MC still stand as important open research devices towards the realization of IoNT. According to [61], issues in the literature. 3-bit barcode-coded DNA strands with dumbbell hairpins can 1) Information Carrying Molecules: Reliable and robust be detected through nanopores with 94% accuracy. In [66], MC necessitates chemically stable IMs that can be selectively four different RNA molecules having different orientations received at the receiver. In addition, environmental factors, are translocated with more than 90% accuracy while passing such as molecular deformation due to enzymes and changes through transmembrane protein nanopores. Nanopore-based in pH can have severe effects on IMs [58] and degrade the detection can also pave the way for detecting cancer biomark- performance of MC. Next, we provide a discussion on several ers from RNA molecules for early detection of cancers as in types of IMs that can be utilized for MC. [69]. However, this requires high sensitivity and selectivity. a) Nucleic Acids: Nucleic acids, i.e., deoxyribonucleic Translocation time of DNA/RNA molecules through acid (DNA) and ribonucleic acid (RNA), are promising can- nanopores depends on the voltage, concentration and length didates as IMsby being biocompatible and chemically stable of the DNA/RNA symbols, and the translocation of symbols especially in in-body applications. In the nature, DNA is carry- can take from a few ms up to hundred ms time frames ing information one generation to another. RNA is utilized as [61], [65]. Considering the slow diffusion channel in MC, messenger molecules for intercellular communication in plants transmission/detection of DNA-encoded symbols do not intro- and animals to catalyze biological reactions such as localized duce a bottleneck and multiple detections can be performed protein synthesis and neuronal growth [59], [60]. In a similar during each symbol transmission. Therefore, the utilization manner, nucleic acids can be exploited in MC-Tx to carry of DNA/RNA strands is promising for high capacity com- information. DNA has a double stranded structure whereas munication between nanomachines as the number of symbols RNA is single stranded molecule often folded on to itself, and in the modulation scheme can be increased by exploiting 7 multiple properties of DNA/RNA at the same time. Hence, within exosomes, vesicles secreted from a multitude of cell the utilization of DNA as information carrying molecule paves types via exocytosis [77]. Artificial transmission of proteins the way for high capacity links between nanomachines by have been long considered within the context of applications enabling higher number of molecules that can be selectively such as protein therapy [78]. From MC point of view, use of received at the Rx. proteins as IMs by genetically engineered bacteria is regarded In addition, information can be encoded into the base as one of the possible biological MC architectures, where sequences of DNAs, which is also known as nucleotide shift- engineered genetic circuits have been proposed to implement keying (NSK). In [22], 35 distinct data files over 200MB various logic gates and operations necessary for networking were encoded and stored by using more than 13 million [79]. nucleotides. More importantly, this work proposes a method e) Other Molecules: Many other types of molecules for reading the stored data in DNA sequences using a random- have been used or considered as IMs in literature. These in- access approach. Thanks to high information density of DNA, clude synthetic pharmaceuticals [80], therapeutic nanoparticles application of NSK in DNA-based MC may boost the typically [81], as well as organic hydrofluorocarbons [20] and isomers low data rates of MC up to the extend of competing with [82]. traditional wireless communication standards. In NSK, infor- 2) Nanomaterial-based MC-Tx Architectures: In the liter- mation carrying DNA and RNA strands can be placed into ature, MC-Tx is generally assumed to be an ideal point source bacteria and viruses. In [47], bacteria based nanonetworks, capable of perfectly transmitting molecular messages encoded where bacteria are utilized as a carrier of IMs, have been in the number, type or release time of molecules to the channel proposed and analytically analyzed. In [70], a digital movie is instantly or continuously, neglecting the stochasticity in the encoded into DNA sequences and these strands are placed into molecule generation process, and the effect of Tx geometry bacteria. However, DNA/RNA reading/writing speed and cost and channel feedback. Despite of the various studies investi- at the moment limit the utilization of NSK in a practical sys- gating MC, physical implementation of MC-Tx stands as an tem. Theoretical and experimental investigation of DNA/RNA- important open research issue especially in micro/nanoscale. encoded MC stands as a significant open research issue in the However, there exist some works on macroscale demonstration MC literature. of MC. In [83], the authors implemented a macroscale MC sys- b) Elemental Ions: Elemental ion concentrations, e.g., tem with an electronically controlled spray as MC-Tx, which Na+, K+, Ca++, dictate many processes in biological sys- is capable of releasing alcohol, and alcohol metal-oxide sensor tems. For instance, in a neuron at steady-state higher Na+ as MC-Rx. According to this experiment, the macroscale and K+ ions are maintained via active transmembrane ion MC setup achieves 0.2bps with 2m communication range. channels in extra- and intracellular medium, respectively, Since there are almost no micro/nanoscale implementation and an action potential is instigated by a surge of Na+ of MC-Tx, we investigate and propose MC-Tx architectures ions into the cell upon activation of transmembrane recep- by exploiting recent advancements in nanotechnology, novel tors by neurotransmitters from other neurons. On the other materials and microfluidics. hand, Ca++ ions are utilized in many complex signaling As discussed in Section II-A, design of MC-Tx in mi- mechanisms in biology including neuronal transmission in cro/nanoscale is an extremely challenging task including vari- an excitatory synapse, exocytosis, cellular motility, apoptosis ous requirements such as biocompability, miniaturization and and transcription [71]. This renders elemental ions one of the lifetime of the device. For reducing the dimensions of MC- viable ways to communicate with living organisms for possible Tx architectures to microscales, microfluidics and microfluidic applications. Correspondingly, many transmitter device [72] droplet technologies are promising. Inside droplets, IMs can and nanonetwork architectures [73] based on elemental ion be transmitted in a precisely controlled manner, such that signaling have been proposed in literature. even logic gates can be implemented with microfluidic chips c) Neurotransmitters: Neural interfacing is one of the as demonstrated in [84]. Feasibility of utilizing droplets for hottest contemporary research topics with a diverse range communication purposes has been suggested in [85]. In ad- of applications. In this respect, stimulation of neurons using dition, microfluidic chips can be fabricated by using Poly- their own language, i.e., neurotransmitters, stands out as the dimethylsiloxane (PDMS), which is a biocompatible polymer best practice. Various transmitter architectures for the most [86]. The authors of [87] theoretically compare achievable data common neurotransmitters, e.g., glutamate, γ-amino butyric rates of passive transport, i.e., diffusive channel, and active acid (GABA), aspartate and acetylcholine have been reported transport, i.e., flow-assisted channel via external pressure or in literature [43], [74], [75]. molecular motors, in a microfluidic environment. According d) Proteins: Proteins comprise the basic building blocks to this study, active transport improves achievable data rates of all mechanisms in life. They are synthesized by cells from thanks to faster movement of IMs from Tx to Rx compared to amino acids utilizing information encoded inside DNA, and passive transport. A new modulation scheme based on distance regulate nearly all processes within the cell, including the pro- between droplets has been introduced in [88] by exploiting tein synthesis process itself, so that they form a self regulatory hydrodynamic microfluidic effects. network. Proteins are commonly used as IMs by biological Although it is not originally proposed as MC-Tx, microflu- systems both in intracellular pathways, e.g., enzymes within idic neural interfaces with chemical stimulation capabilities, vesicles between the endoplasmic reticulum and the Golgi i.e., devices that release neurotransmitters such as glutamate apparatus [76], and intercellular pathways, e.g., hormones or γ-aminobutyric acid (GABA) to stimulate or inhibiting 8

(a) (b) (c) Transmitted (d) Transmitted IMs IMs Hydrophobic Nanopore Nanoporous Nanoporous graphene graphene membrane membrane

Liquid with high Liquid with high IM concentration IM concentration Heating electrode

Reservoir Reservoir Nanoporous walls walls Reservoir Reservoir graphene walls Electrode walls Molecule Electrode Electrical stimuli membrane wax responsive hydrogel Fig. 4: MC-Tx Architectures (a) microfluidic-based MC-Tx with hydrophobic nanopore, (b) microfluidic-based MC-Tx with hydrophobic nanopore and nanoporous graphene membrane, (c) MC-Tx with thin film hydrogels with nanoporous graphene membrane, (d) MC-Tx with molecule wax and nanoporous graphene membrane. neural signals [15], [16], [89], operates same as MC-Tx. Up to this point, we consider the design of MC-Tx only Therefore, the studies on neural interfaces with chemical in liquid medium. For airborne MC, we propose an MC- stimulation capabilities can be considered as baseline for Tx architecture consisting of a molecular reservoir sealed by designing micro/nanoscale MC-Tx. However, leakage of IM a porous graphene membrane, accommodating a wax layer molecules while there is no signal transmission stands as containing IMs as in Figure 4(d). The rest of the reservoir is a significant challenge for microfluidic based MC-Txs. It is filled with water, in which the IMs dissolve. Upon application not possible to eliminate the problem, but there are some of heat via E-field through conducting walls of the reservoir, possible solutions to reduce or control the amount of leakage. the module sweats IM-rich water through membrane pores, Hydrophobic nanopores can be utilized as shown in Figure and IMs become airborne upon evaporation, and provide a 4(a) such that the liquid inside the container can be separated continuous release of IMs. from the medium when there is no pressure inside the MC- 3) Biological MC-Tx Architectures: An alternative ap- Tx. This solution has two drawbacks. Since there is a need for proach to nanomaterial-based architectures for MC, which in external pressure source, it is hard to design a practical stand- most cases are inspired by their biological counterparts, resides alone MC-Tx with this setup. Secondly, this solution does not in rewiring the already established molecular machinery of completely eliminate the IM leakage; hence, the authors of biological realm to engineer biological nanomachines that [90] suggest the utilization of porous membranes and electrical network via MC to accomplish specific tasks. At the cost of control of fluids to further improve MC-Tx against IM leakage. increased complexity, this approach has various advantages Nanoporous graphene membranes can provide molecule over nanomaterial-based designs including inherent biocom- selectivity [91] by adjusting pore size depending on the size of patibility and already integrated production, transport and IM. In addition, graphene is also proven to be biocompatible transmission modules for a wealthy selection of IMs and as demonstrated in [92]. If the pore size of the graphene architectures. membrane can be adjusted in a way that IMs can barely a) Bacterial Conjugation-based Transmission: Bacte- pass through, negatively charging the graphene membrane can rial conjugation is one of the lateral gene transfer processes decrease the pore size with the additional electrons such that between two bacteria. More specifically, some plasmids in- some level of control can be introduced to reduce lM leakage. side bacteria, mostly circular small double-stranded DNA In addition, an electrode plate placed at the bottom of the molecules physically distinct from the main bacterial DNA, container can be utilized to pull and push charged IMs as can replicate and transfer itself into a new bacteria [96]. illustrated in Figure 4(b). This way, an electric field (E-field) These plasmids encode the conjugative ”sex” pilus, which can be generated in the liquid containing charged IMs and upon receiving a right molecular stimulus from a neighboring the direction of E-field can be utilized to ease or harden the bacteria is translated. The produced pilus extends out of the release of charged IMs. donor cell, attaches to the recipient cell, and then retracts to get MC-Tx can be also realized with electrical stimuli- the two cells in contact with their intracellular media joined responsive thin film hydrogels by performing E-field mod- through the pilus hole. Single strand DNA (ssDNA) of the ulated release and uptake of IMs [14] as in Figure 4(c). plasmid is then transferred from the donor cell to the recipient Hydrogels, widely utilized in smart drug delivery, are bio- cell. utilizing bacterial conjugation as a means of information compatible polymers that can host molecules, and reversibly transmission for MC necessitates at least partial control over swell/deswell in water solution upon application of a stimuli, bacteria behavior, which is achieved by means of genetically resulting in release/uptake of molecules. This architecture engineering the bacteria. can be further improved via a porous graphene controlling At the core of all genetical engineering schemes lies the its micro-environment. E-field stimuli can be generated by concept of gene regulation via a process known as RNA two electrodes placed at the opposite walls of the reservoir. interference (RNAi), which has provided us with the means Refilling the IM reservoirs is another significant challenge in of manipulating gene expressions in targeted cells or bacteria, MC-Tx design, which can be solved with the replenishable paving the way for genetical engineered bacteria based MC drug delivery methods, e.g., oligodeoxynucleotides (ODN) architectures [97]. RNAi is observed to serve as a mediator modification [93], click chemistry [94], and refill lines [95]. of inter-kingdom MC [98]. In particular, it is shown that 9 short hairpin RNA (shRNA) expressing bacteria elicit RNAi for varying system parameters such as quantity of bacteria, in mammals [99], rendering bacteria mediated RNAi-based area of deployment and number of target sites. An important diagnosis and therapeutics a promising prospect [100]. In this factor that restricts reliable successful delivery in conjugation respect, recently [101] proposed a platform of genetically based bacterial networks is incomplete DNA transfer due engineered bacteria as vehicles for localized delivery of ther- fragile nature of the process, the effect of which is pronounced apeutics towards applications for Crohn’s disease. From MC over multiple transfers. To mitigate losses due to this effect, perspective, genetically engineered bacteria have been envi- which always results in a loss of information from the tail sioned to be utilized in various ways with different transmitter part of DNA, [108] devises a Forward-Reverse Coding (FRC) architectures. Below, we collate various novel research direc- scheme, where messages are encoded in both directions and tions in biological transmitter architectures for MC enabled by sent simultaneously over the same channel via dedicated sets the recent advancements in genetic engineering. of bacteria, to equally distribute losses to both ends of DNA. It is important to note that, conjugative pili are typically The performance of FRC is later compared with a cyclic shift few microns in length, which from MC point of view dramat- coding (CSC) scheme, in which DNA message is partitioned ically decreases the transmission radius. Accordingly, [47] has into N smaller blocks and is cyclically shifted to create proposed engineered bacteria with flagella, e.g., E. coli as the N versions of the same content starting with corresponding carriers of information, i.e., sequenced DNA strands, which blocks, and similar to FRC transmitted simultaneously via establish a communication link between two nanomachine dedicated bacteria populations [109]. Expectedly, CSC pro- nodes that can interface and exchange DNA strands with vided higher link probability than FRC, which outperformed the bacteria. The sequenced DNA message resides within a straight encoding. In a very similar spirit, [110] considers plasmid. The behavior of bacteria is controlled via chemotaxis packet fragmentation at transmitter and reassembly at receiver, by release of attractants from the nodes, and is regulated by which has the additional advantage of faster propagation encoded active regions on the plasmid. To avoid interference due to diminished size of carriers. They observe that this with behavior regulation the message section of the plasmid is protocol increases reliability over diffusive channels, however, inactivated, i.e., it is not expressed, which can be achieved by it degrades communication when there is strong enough drift, avoiding consensus promoter sequences within the message. which can be attributed to the heightened diffusion rates. Consensus promoters are necessary for the RNA polymerase to The MC transmitter module of the architecture based on attach to the DNA and start transcription. The message section bacterial conjugation outlined above, indeed, is composed of the plasmid also contains the destination address, which of several submodules, i.e., the transmitter module of the renders message relaying across nodes possible. The authors nanomachine node, the bacteria itself via chemotaxis and the develop a simulator for flagellated bacteria propagation, which pilus-based sexual conjugation module, which is a reflection they combine with analytical models for biological processes of the trademark high complexity involved with biological involved to obtain the end-to-end delay and capacity of the architectures. In return, the reward is the achievement of un- proposed MC channel for model networking tasks. In [52], precedented data rates via MC thanks to the high information utilization of flagellated bacteria for medium range (µm- density of DNA. mm) MC networks is suggested, where authors present a b) Virus-based Transmission: Viruses have initially physical channel characterization of the setup together with been studied as infectious agents and as tools for investigating a simulator based on it. [102] extends the model in [47] by cell biology, however, their use as templates for transferring accounting for mutations in the bacteria population, and also genetic materials to cells has provided us with the means of considering asynchronous mode of operation of the nodes. genetically engineering bacteria [111], and unlocked a novel In [103], the authors also consider a similar setup utiliz- treatment technique in medicine, e.g., gene therapy [112]. A ing bacterial conjugation, but employ opportunistic routing, virus is a bio-molecular complex that carries DNA (or RNA), where opportunistic conjugation between bacteria is allowed which is packed inside a protein shell, called capsid, that is in contrast to strict bacteria-node conjugation assumed in [47]. enveloped by a lipid layer. The capsid has (at least) an entrance However, it requires node labeling of bacteria, and additional to its interior, through which the nucleic payload is packed via attractant release by bacteria to facilitate bacteria-bacteria a ring ATPase motor protein [113]. Located near the entrance contact for opportunistic routing. The authors extend this work are functional groups that facilitate docking on a cell by acting with [104] by additionally assuming the nanomachine nodes as as ligands to receptors on it. Once docked, the nucleic content capable of releasing antibiotics that kill bacteria with useless is injected into the cell, which triggers the cell’s production or no content to decrease noise levels by avoiding overpop- line to produce more of virus’ constituent parts and DNA. ulation. Moreover, they allow multiple number of plasmids These self-organize into fully structured viruses, which finally per bacterium, which, contrary to their previous work and burst out of the host cell to target new ones. [47], allows simultaneous communication between multiple From MC perspective, viral vectors are considered to be source and destination nodes over the same network. Later, one of the possible solutions in transmitting DNA between [105] considers the design of the nanomachine node in more nanonetworking agents. The concept is similar to bacterial detail, and [106] derives an analytical model to estimate the conjugation based transmission, as both encode information successful delivery rates within proposed framework. [107] into transmitted DNA, but, in contrast viral vectors are im- analyzes via simulations the end-to-end delay and reliability motile and their propagation is dictated by passive diffusion. of conjugation based transmission within the same scenario However, they are comparatively smaller than bacteria, so 10 that more of them can be deployed in a given channel, sensing as a tool for power amplification of MC signals and they diffuse faster. Moreover, the ligand-receptor docking increasing the range of transmitted signals. In contrast, [119] mechanism, which serves as a header for receptor/cell specific considers exosome secretion as a possible means for real- long range targeting, enables the possibility of design of very ization of nanonetworks composed of large number of bio- complex and large scale networking schemes with applications nanomachines. [31] presents a detailed model of engineered in gene therapy [4]. Furthermore, considering high information genetic circuitry based on mass action laws that regulate density of DNA, virus based MC stands out as one of the MC gene expressions, which in turn dictate transmitter protein protocols viable to support high data rates. Yet, models of MC production. Their work stands as a basis for future engineered networks utilizing viral vectors are still very few. In particular, genetic circuitry based protein transmission. [79] proposes utilization of engineered cells as platforms d) Enzyme Regulated Ca++ Circuits: Ca++ ions are for devices and sensors to interface to nanonetworks. These utilized in many complex signaling mechanisms in biology engineered cells are assumed to be capable of virus production including neuronal transmission in an excitatory synapse, and excretion to facilitate desired networking, where a modular exocytosis, cellular motility, apoptosis and transcription [71]. approach is presented for modeling of the genetic circuitry Moreover, they play a major role in intracellular and inter- involved in the modulation of viral expression based on incom- cellular signal transduction pathways, where the information ing extracellular signaling. Based on the model developed in is encoded into local Ca++ concentration waves under the [79], [114] and [115] investigate viral MC networking between control of enzymatic processes. Intracellular organelles in- nanomachines that communicate with each other in a diffusive cluding mitochondria and endoplasmic reticulum accumulate medium via DNA messages transmitted within viruses, where excess Ca++ ions, and serve as Ca++ storages. Certain the former analyses the reliability of a multi-path topology, cellular events, such as an extracellular signal generated by and the latter derives reliability and delay in multi-hop relay a toxin, trigger enzymatic processes that release bound Ca++ networks. from organelles effectively increasing local cytosolic Ca++ c) Genetic Circuit Regulated Protein Transmission: We concentrations [126]. These local concentrations can propagate have so far investigated biological transmitter architectures, like waves within cells, and can be injected to adjacent cells where the message to be conveyed has been encoded in through transmembrane protein gap junction channels called sequences of nucleotides, e.g., DNA or RNA. Yet, the most connexins [127]. abundant form of intercellular MC interaction in nature occurs Establishing controlled MC using this inherent signaling via proteins, whose expression levels are determined by the mechanism was first suggested by [128], where authors con- genetic circuitry and metabolic state within the cells. Typically, sider communicating information between two nanomachines proteins that are produced within a cell via transcriptional pro- over a densely packed array of cells that are interconnected cesses are either excreted out via specialized transmembrane via connexins. The authors simulate MC within this channel protein channels, or packed into vesicles and transported to to analyze system parameter dependent behavior of inter- extracellular medium via exocytosis. As a result of exocytosis, cellular signal propagation and its failure. They also report either the vesicle is transported out wholly, e.g., an exosome on experiments relating to so-called cell wires, where an [116], or it fuses with the cell membrane and only the contents array of gap junction transfected cells are confined in a wire are spewed out. Proteins on the membranes of exosomes configuration and signals along the wire are propagated via provide addressing via ligand-receptor interactions with mem- Ca++ ions. The communication was characterized as limited brane proteins of recipient cell. Upon a match, the membranes range and slow speed. Later, Ca++ relay signaling over 1 of exosome and the recipient cell merge, and exosome contents cell thick cell wires were investigated in [79] and [129] via enter recipient cell. This establishes a one-to-one MC link dedicated simulations, where the former explored amplitude between two cells. In case the contents are merely spewed out and frequency modulation characteristics and the latter aimed to the external medium, they diffuse around contributing to the at understanding the communication capacity of the channel overall concentrations within the extracellular medium. This under stochastic effects. Simulations to determine the effects corresponds to local message broadcast, and it has been long of tissue deformation on Ca++ propagation and the capacity observed that populations of bacteria regulate their behavior of MC between two nanomachines embedded within two- according to resulting local molecule concentrations resulting dimensional cell wires with thickness of multiple cells were from these broadcast messages, referred to as the phenomenon carried out in [130]. As part of their study, authors propose var- of quorum sensing. ious transmission protocols and compare their performance in This mode of communication, even though lacking the terms of achieved rates. In a later study [131], authors employ information density of nucleotide chains, and therefore sup- Ca++ signaling nanomachines embedded within deformable porting lower data rates, are commonly employed by nature cell arrays to infer deformation status of the array as model for as MC schemes. They are considerably more energy efficient tissue deformation detection. This is achieved by estimating compared to DNA transmission schemes, which justifies their the distance between nanomachines from observed information use by nature as signaling agents for comparatively simple metrics coupled with strategic placements of nanomachines. nanonetworking tasks. In this respect, [117] proposes quorum Motivated by tissue health inference via embedded nanoma- sensing as a means to achieve synchronization amongst bacte- chines, the authors of [132] identify three categories of cells ria nodes of a nanonetwork, where they model a bacterium that employ Ca++ signaling, namely excitable, non-excitable as a finite state automaton, and [118] investigates quorum and hybrid, which respectively model muscle cells, epithe- 11

TABLE I: Comparison Matrix for MC Modulation Schemes

Description Encoding Mechanism ISI Reduction # of Molecule Type # of Symbols

On-off keying (OOK) [120] Concentration No 1 2 Concentration shift keying (CSK) [20] Concentration No 1 (b Concentration levels) 2b Pulse amplitude modulation (PAM) [121] Concentration No 1 (b Concentration levels) 2b Molecule shift keying (MoSK) [10] Molecule type Moderate k k Depleted MoSK (D-MoSK) [122] Molecule type No k 2k Isomer-based ratio shift keying (IRSK) [82] Molecule ratio Moderate k k Release time shift keying (RTSK) [21], [123], [124] Release timing No 1 (b Timing intervals) 2b Molecular array-based communication (MARCO) [125] Molecule order High k 2k lium cells and astrocytes, and model Ca++ communication molecule release time as in Table I. The first and simplest behavior within channels comprised of these cells. We refer modulation method that was proposed for MC is on-off keying the reader to [133] for a more detailed discussion of existing (OOK), in which certain number of molecules are released for literature, theoretical models, experiments, applications and high logic and no molecule is released to represent low logic future directions in this field of MC. [120]. In a similar manner by using a single molecule, concen- e) Other Biological Architectures: In addition to the tration shift keying (CSK) that is analogous to amplitude shift biological transmission architectures described above, molec- keying (ASK) in traditional wireless channels is introduced in ular motors sliding on cytoskeletal protein structures, e.g., order to increase the number of symbols in the modulation microtubules, and carrying cargo, i.e., vesicles, between cells scheme by encoding information into concentration levels is another option that has been considered by the MC commu- [20]. In [121], a similar modulation scheme based on the nity. In particular, [134] and [135] both describe a high level concentration levels is proposed and named as pulse amplitude architecture design for MC over such channels. The compar- modulation (PAM), which uses pulses of continuous IM release ison of active, i.e., molecular motors on microtubules, and instead of instantaneous release as in CSK. The number of passive, i.e., diffusion, vesicle exchange among cells shows concentration levels that can be exploited significantly depends that, active transport is a better option for intercellular MC in on molecule type and channel characteristics as ISI at MC-Rx case of low number of available vesicles, and passive transport can be a limiting factor. can support higher rates when large numbers of vesicles are Hitherto, we have discussed modulation techniques that available[136]. Two design options to form a microtubule use single type of molecules. However, information can be nanonetwork in a self-organizing manner, i.e., via a polymer- encoded by using multiple molecules such that each molecule ization/depolymerization process and molecular motor assisted represents different symbol, and k information symbols can be organization, are proposed in [51]. A complementary approach represented with 2k different types of molecules in molecule to molecular motor based microtubular MC is presented in shift keying (MoSK) [10]. That is, 1-bit MoSK requires two [137], where an on-chip MC testbed design based on kinesin molecules to encode bit-0 with molecule A and bit-1 with molecular motors is presented. In this approach, instead of molecule B. The modulation of each molecule in MoSK is molecular motors gliding over microtubules as carriers of based on other modulation techniques such as OOK. MoSK molecules, microtubules are the carriers of molecules, e.g., can achieve higher capacity, but the main limiting factor for ssDNA, gliding over a kinesin covered substrate. this modulation type is the number of molecules that can Other transmitter approaches that involve the use of biolog- be selectively received. The authors of [122] further improve ical entities are biological-nanomaterial hybrid approaches. A MoSK by representing low logic with no molecule release and promising approach is to utilize IM production mechanisms of enabling simultaneous release of different type of molecules bacteria in nanomaterial based transmitter architectures. In this such that 2k symbols can be represented with k molecules, direction, [138] reports on an optogenetically controlled living which is named as depleted MoSK (D-MoSK). That is, 1- hydrogel, that is, a permeable hydrogel matrix embedded with bit D-MoSK is equivalent to OOK. 2-bit D-MoSK requires bacteria from an endotoxin-free E. coli strain, which release only two molecules, and four distinct symbols can be encoded IMs, i.e., anti-microbial and anti-tumoral drug deoxyviolacein, with these molecules (molecule A and molecule B), such as in a light-regulated manner. The hydrogel matrix is permeable N (00), A (01), B (10), and AB (11), where N represents to deoxyviolacein, however spatially restricts movement of no molecule release. Furthermore, the authors of [82] propose the bacteria. This hybrid approach proposes a solution the the utilization of isomers, i.e., the molecules having the same the reservoir problem of nanomaterial based architectures. atoms in a different orientation, and a new modulation scheme, Moreover, to cover a variety of IMs simultaneously this named as isomer-based ratio shift keying (IRSK), in which approach can be built upon by utilizing many engineered information is encoded into the ratio of isomers, i.e., molecule bacteria, and controlling their states via external stimulus to ratio-keying. control their molecular output [139]. The release time of molecules can be also used to encode information. In [121], the authors propose pulse position C. Modulation Techniques for MC modulation in which signaling period is divided into two In MC, several modulation schemes have been proposed blocks such that a pulse in the first block means high logic and to encode information into concentration, molecule type, and a pulse in the second block means low logic. More complex 12 modulation schemes-based on release timing, i.e., release time to ISI. Moreover, as coding has a computational burden on shift keying (RTSK), where information is encoded into the both transmitter and receiver ends, and energy is a scarce time interval between molecule release has been investigated resource at nanoscale, energy efficiency of employed channel in [21], [123], [124], [140]. The channel characteristics in codes is also a crucially important aspect. This calls for case of RTSK is significantly different than other modulation utilization lower complexity block codes, such as simple parity schemes as additive noise is distributed with inverse Gaussian codes or cyclic codes, e.g., Hamming codes [24], instead of distribution in the presence of flow in the channel [21], and the state-of-the-art high complexity codes with high compu- Levy distribution without any flow [124]. tational burden like Turbo codes [23]. On the other hand, In MC, ISI is an important performance degrading factor the noisy nature of MC channels and over-pronounced effects during detection due to random motion of particles in the of ISI renders channel codes developed for conventional EM diffusive channels. The effects of ISI can be compensated communications ill-adjusted for MC, calling for invention of by considering ISI-robust modulation schemes. In [141], an novel coding techniques specifically tailored for MC. Table II adaptive modulation technique exploiting memory of the enlists the channel coding practices so far employed in MC channel is utilized to encode information into emission rate literature. Below, we summarize these works and highlight of IMs, and this approach makes the channel more robust their contributions, where we focus more on codes that have against ISI by adaptive control of the number of released been specifically designed for MC. molecules. In addition, the order of molecules can be also The first MC specific code in literature aimed at mitigating used for information encoding as in molecular array-based the effects of ISI in MC, as it is the main source of high communication (MARCO) [125]. In this approach, different BERs. In [145] authors introduce the ISI-free coding scheme types of molecules are released consecutively to transmit under MoSK modulation, where two distinguishable molecules symbols, and by assuming perfect molecular selectivity at the encode for bit 0 and bit 1, respectively. The receiver is transmitter, the effects of ISI can be reduced. absorbing, i.e., detects everything that hits, and it immedi- The modulation schemes based on concentration, molecule ately receives bit 0 or bit 1 upon detection depending on type/ratio/order, and molecule release time offer limited num- the type of detected molecule. The authors work with the ber of symbols. Therefore, MC suffers from low data rates example of a (4, 2, 1) ISI-free code, where an (n, k, l) ISI- by considering limited number of symbols and slow diffusive free code is an (n, k) block code, i.e., maps k-bit information propagation. To tackle this problem, large amount of informa- into n-bit codewords, and is error-free provided that there tion can be encoded into the base sequences of DNAs, i.e., are no more than level-l crossovers. Here, crossover is the Nucleotide Shift-Keying (NSK). For this purpose, information phenomenon of late detection of a molecule belonging to can be encoded directly using nucleotides with an error coding previously transmitted symbols, and a level-l crossover means algorithm such as Reed Solomon (RS) block codes [142], or an that the detected molecule was transmitted l symbols ago. alphabet can be generated out of DNA sequences (100-150bp ISI-free code is a fixed code, in that, it is invariant with per letter) to encode information. The latter approach can yield respect to change in channel parameters. The (4,2,1) code higher performance in terms of BER considering the complex- implemented in [145] is based on the idea of finding a ity and size of MC-Tx and MC-Rx architectures. Although codebook with codewords, whose level-1 permutation sets, identification of base pairs with nanopores can be performed i.e., possible detection sequences under maximum level-1 with relatively low costs and high speeds [143], [144], there crossover assumption, are disjoint. However, level-1 permuta- is yet no practical system to write DNA sequences with a mi- tion sets of codewords depend on values of neighboring bits at croscale device. Therefore, future technological advancements the boundary of contiguous codewords, where if they are same, towards low-cost and practical synthesis/sequencing of DNA crossovers between contiguous codewords do not contribute are imperative for MC communications with high data rates, new elements to the level-1 permutation set, making finding e.g., on the order of Mbps. codewords with disjoint permutation sets easier. To achieve this, authors devise a two state encoder architecture, whose codeword assignments and state diagram are illustrated in Fig. D. Coding Techniques for MC 5 together with an example encoding. Note that contiguous Encoding of information before transmission is classically codewords have neighboring bits always same. The receiver done for two reasons, source coding is done for statistically decodes the information bits from the codeword received by efficient representation of data form a discrete input source, adding the number of 1’s modulo n = 4, and converting and channel coding is done to control errors that occur due to the result to binary, which is a fairly simple decoding rule, channel noise via introducing redundant bits. Source coding and therefore favorable for MC. The authors also compare practices are independent of channel characteristics, and as a the ISI-free (4, 2, 1) code with convolutional and repetition result they do not differ for MC with respect to traditional codes, and verify comparable BER performance with much communications from ICT point of view. For this reason, less computational burden. In their following work [146], we do not cover source coding in this review. On the other authors extend the ISI-free (n, k, l) codes to account for higher hand, MC channels are typically diffusive, a process which level crossovers, namely, for levels l = 2, 3, 4, 5. Furthermore, has slow and omnidirectional propagation. As a consequence they introduce the ISI-free (n, k, l, s) codes, in which the IMs quickly accumulate in the channel after a series of codewords have at least l final and s initial identical bits, transmissions, rendering MC extremely noisy and susceptible instead of the symmetric at least l identical bits at both ends of 13

TABLE II: Comparison Matrix for MC Channel Codes

Transmission/ Channel Model # of Mol. Code Mod. Detection Reception C.D. Coding Rate ( k ) E.C. # of Mol. Prop./Dim./B.N. Types n

2 2 4 3 3 4 5 ISI-free code [145], [146] MoSK Reg./Single Diff.+Drift/1d/No Abs. Imm. No 2 4 , 5 , 7 , 8 , 11 , 23 , 47 No MoCo-based code [147] OOK Reg./Single Diff.+Drift/1d/Yes Abs. Add.Thr.=1 Yes 1 2/4 No 2m−m−1 Hamming codes [148], [149] OOK Reg./Many Diff./3d/No Non-Abs. Add.Thr. No 1 2m−1 , m = 3, 4, 5 Yes SOCC [150] OOK Reg./Many Diff./3d/No Non-Abs. Add.Thr. No 1 k/(k + 1), k = 1, 2 Yes SOCC [151] OOK Reg./Many Diff./3d/No Abs. Add.Thr. No 1 k/(k + 1), k = 1, 2 Yes 4s−3s EG-LDPC [152] OOK Reg./Many Diff./3d/No Abs. Add.Thr. No 1 4s−1 , s = 2, 3, 4 Yes Convolutional code [153] PAM Reg./Many Diff./3d/No Abs. Add.Thr. No 1 1/2 No SPC code [154] OOK Reg./Many Diff./3d/Yes Reactive LLR No 2 2/3 No Abbreviations — Abs.: Absorbing – Add.Thr.: Additive with threshold – B.N.: Background Noise – C.D.: Channel Dependent – Diff.: Diffusion – Dim.: Dimension – E.C.: Energy Considered – Imm.: Immediate – Mod.: Modulation – Mol.: Molecule – Non-Abs.: Non-absorbing – Prop.: Propagation – Reg.: Regular

ISI-free (n, k, l) codes, and show that they significantly out- perform (n, k, l) codes under similar computational burdens. (a) Information Starting Level-1 Starting Level-1 bits with 0 permutation with 1 permutation In essence, the motivation for (n, k, l, s) codes comes from 00 0000 0000 1111 1111 the asymmetry of inter-codeword error probabilities arising 01 0001 0001, 0010 1000 0100, 1000 from crossover at the start of a codeword and at the end of a 10 0011 0101, 0011 1100 1100, 1010 codeword. More specifically, if one compares the probability 11 0111 0111, 1011 1110 1101, 1110 of a given molecule having level-l crossover forwards via arriving later than the l molecules released after it, to the (b) 01/0001 probability of having level-l crossover backwards via the 10/0011 preceding l molecules arriving after the given molecule, one 11/0111 00/0000 00/1111 finds latter to fall far more rapidly with increasing l. Thus, 0 11/1110 1 to reduce computational burden s is typically chosen lower 10/1100 than l signifying the low probability of backwards crossover 01/1000 errors. Finally, the authors demonstrate that ISI-free (n, k, l, s) codes can deliver better BERs than convolutional codes with (c) Uncoded less computational resources. As its weaknesses, the work Sequence 1 0 0 0 0 1 1 1 considers a very simple one dimensional diffusive channel Encoded model with positive drift velocity, which lacks many phenom- Sequence ... 0 0 0 1 1 1 1 1 1 1 0 0 0 0 1 1 1 1 ... ena diffusive MC enjoys in three dimensions. In particular, transmitted IMs are doomed to hit the receiver, which is very different from the three dimensional case, where there Fig. 5: (a) Employed codewords for two states, e.g., starting is always the probability that no molecules will reach the with 0 or 1, together with their level-1 permutations, (b) the receiver. The extent of this simplification reveals itself in the state transition diagram, and (c) an encoding example for the assumption that the transmitter releases a single molecule per ISI-free (4, 2, 1) code. Note the same bits at the boundaries of symbol, which, thanks to the drift in channel and the absorbing contiguous words. [145] nature of receiver, is always detected.

MC-adapted version of classical Hamming codes were introduced in [147], where the traditional Hamming distance codes, the authors demonstrate that the code generated using metric on the codeword space, given by the number of bit MoCo performs superior to Hamming code by carrying out an differences between two binary codewords, is replaced by the error rate analysis for both codes. However, as shortcomings, so-called molecular coding distance function (MoCo). In its MoCo depends on detection probabilities that are sensitive to essence, MoCo is defined in terms of the negative logarithms variations in channel properties, and are in general unknown of probabilities P r({x → y}) of receiving codeword y when to Tx and Rx. Moreover, even if adaptive techniques may be x was transmitted, and the code aims at generating a codebook envisioned, calculation of MoCo-based codebook (at Tx) and with maximal minimum pairwise MoCo distance between decoding region partition (at Rx) require significant compu- constituent codewords. MoCo distance is not a metric, as it tational resources at Tx and Rx, respectively, and overhead is not symmetric, and the triangular inequality is not verified communication would have to be considered for synchronized by the authors. This work, too, considers a one dimensional code updating. diffusive channel with drift and an absorbing receiver, how- The authors of [148] propose classical Hamming codes ever, in contrast to [145], [146], it uses synchronized time to introduce error correction in MC, where they consider a slotted OOK modulation scheme with only a single type of three-dimensional diffusive channel with time slotted OOK IM and the receiver is additive with a threshold equal to modulation scheme in a channel with finite memory. The 1, i.e., it counts the number of hits in a period and claims receiver is modeled as a non-absorbing sphere which im- high logic reception with a single hit. In case of (4, 2) block mediately detects molecules that arrive to it, and reception 14 is decided upon additive count of arriving molecules during coding with high transmission rate and M = 1, i.e., OOK transmission period. Hamming codes are error correcting block modulation, does outperform the uncoded transmission in codes with coding ratio k/n = (2m −1)/(2m −m−1), where short- and medium-range communications, no coding does bet- m is the number of parity check bits, and [148] considers ter than convolutional codes in long range MC. Furthermore, Hamming codes for m = 3, 4, 5. Their results show that increase in the number of pulse amplitude levels causes de- Hamming codes can deliver coding gains up to ≈ 1.7dB terioration in achieved BERs, which is attributed to increased at transmission distances of 1µm and for low BERs. Here, ISI, implying that, OOK modulation is better suited to MC coding gain is defined as the gain the code introduces in than PAM. required number of IMs per transmission to achieve a given All the aforementioned works apply various channel coding BER. At high BERs, i.e., low quantities of transmitted IMs, techniques for error correction in MC, however they do not extra ISI introduced by parity bits overweighs error correction, provide any details into mechanisms of implementation of and uncoded transmission performs better. The authors also these codes from device architecture perspective. In [155], incorporate an energy model for transmission, where energy authors devise a molecular single parity check (SPC) encoder is taken to be proportional to the number of transmitted with OOK modulation for an MC design based on genetically IMs, and show that the energy required to transmit the extra engineered bacteria, which are assumed to network with each parity bits causes coded transmission to be energy inefficient other via signaling molecules, e.g., N-acyl homoserine lac- at small communication distances, however coding becomes tones (AHLs). The implementation of joint encoder-modulator more efficient for larger distances. Later on, over the same module is achieved via design of genetic circuits that regulate channel model a Hamming minimum energy code (MEC) gene expression levels, and the transmission materializes from scheme was proposed in [149]. In a trade off of having larger ensuing biomolecule concentrations dictated by biochemical codeword lengths against generating codewords with lower av- reactions. Developed SPC encoder, which appends to 2 bits erage weights by using more 0-bits, the authors trade between information a parity check bit via biological XOR gate based rate and energy efficiency of communication. In subsequent on designed genetic circuits, provides an error detection mech- works [150], [151], again over the same channel except with anism, however with no correction. Still, the introduced design an absorbing receiver in [151], self-orthogonal convolutional serves as a basis for genetic circuit based designs of more codes (SOCCs) are proposed, and their performance against complex block codes with error correction capabilities. In this Hamming MECs and uncoded transmissions are investigated paper, there is no evaluation of the proposed coding scheme with respect to both BER and energy efficiency. Both works as the paper considers only the transmitter side of MC. A conclude that, in nanoscale MC SOCCs have higher coding year later, in [154], authors extended their work in [155] by gains, i.e., they are more energy efficient, compared to uncoded introducing biological analog decoder circuit, which computes transmission, and to the Hamming MECs when low BER the a-posteriori log-likelihood ratio (LLR) of transmitted bits (10−5-10−9) region. Moreover, SOCCs are also reported to from observed transmitter concentrations. LLR is defined as have shorter critical distances than Hamming MECs, where the gain of probability of detection over non-detection in dB. the critical distance is defined as the distance at which extra This enabled them to analyze the whole end-to-end MC over energy requirements of employing coding are compensated a diffusive channel. Via simulations they manage to verify by the coding gain. [151] additionally explores the energy intended operation of designed modulated SPC encoder and budget of nano-to-macro and macro-to-nanomachine MC, and the analog decoder, and observe network performance close arrive at the conclusion that in MC involving macro-machines to an electrical network operating in high noise. the critical distance of the codes decrease. Yet in another work [152], Hamming codes are evaluated against cyclic 2 III.MOLECULAR COMMUNICATION RECEIVER dimensional Euclidean geometry low density parity check (EG-LDPC) and cyclic Reed-Muller codes by considering the The MC receiver (MC-Rx) recognizes the arrival of target same channel model as in [151]. Again, the comparison of molecules to its vicinity and detects the information encoded in codes is carried out for different MC scenarios involving nano a physical property of these molecules, such as concentration, and macro-machines, and it reveals in the case of nano-to- type or release time. To this aim, it requires a molecular nanomachine MC that, in BER region 10−3 − 10−6 Hamming receiver antenna that consists of a biorecognition unit fol- codes with m = 4 are superior, and at lower BER regions lowed by a transducer unit. The biorecognition unit, i.e., LDPC codes with s = 2 exhibit the lowest energy cost. Here, the interface with the molecular channel, holds a molecular s ≥ 2 is the density parameter in LDPC codes, where coding recognition event specifically selective to the information density increases to 1 monotonically as s → ∞. Moreover, carrying molecules, e.g., it selectively reacts to these target in macro-to-nano and nano-to-macro MC the results indicate molecules. Then, the transducer unit generates a processable that LDPC codes with s = 2 and s = 3 are the best options, signal, e.g., electrical or biochemical signal, based on this respectively. molecular reaction. Finally, a processing unit is needed to The performance of convolutional coding techniques in detect the transmitted information based on the output of the diffusive MC systems has been investigated by [153] utilizing molecular antenna. The interconnection of these components PAM with M = 1, 2, 4 pulse amplitude levels for varying key in an MC-Rx is illustrated in Fig. 6 [17]. Since this structure is factors such as transmission rate and communication range fundamentally different from EM communication receivers, it (0.8µm-1mm). The findings indicate that, while convolutional is necessary to thoroughly investigate the receiver architecture 15

receptors, i.e., they must return to their initial state after Molecular Channel Receiver signal detection to be ready for the next channel use. • Energy efficiency: Due to limitations of nanomachines, Biorecognition the energy usage of the MC-Rx must be optimized. Addi- Unit tionally, as discussed in Section II-A, batteries may not be feasible solutions for long term activity of nanomachines, Transducer Processing e.g., as an implanted device. Thus, the receiver may need Unit Unit to be designed with EH units to be energy self-reliance. • Biocompatibility and biodurability: One of the most important MC application areas is the life science, e.g., it

T promises diagnosis and treatment techniques for diseases caused by dysfunction of intrabody nanonetworks such as neurodegenerative diseases [3]. These in vivo applications dictate further requirements for the device. Firstly, it must not have any toxic effects on the living system. Moreover, the device needs to be flexible not to cause any injury to the living cells due to mechanical mismatches. Furthermore, there must not be any physiological re- actions between device and environment and it must Information Molecular Receiver not cause immunological rejection. On the other hand, Molecules Antenna the physiological environment should not degrade the performance of the device with time. Fig. 6: Components of an MC-Rx. • Miniaturization: Lastly, to be integrated into a nanoma- chine, the MC-Rx must be built on micro/nanoscale specification. To this aim, in this section, we first discuss the components. requirements of a receiver to be operable in an MC application and the communication theoretical performance metrics that B. Communication Theoretical Performance Metrics must be taken into consideration while designing the receiver. In addition to the general performance metrics defined for Then, we review the available approaches in physical design EM communications, e.g., Signal-to-noise ratio (SNR), Bit of MC-Rxs, which can be categorized into two main groups, error rate (BER), and mutual information, new performance (i) biological receivers based on synthetic gene circuits of metrics are needed to fully evaluate the functionality of an engineered bacteria and (ii) nanomaterial-based artificial MC- MC-Rx since molecules are used as IMs in MC. The most Rx structures. important performance metrics are summarized as follows. • Limit of detection (LoD): LoD is a well-known perfor- A. Design Requirements for MC-Rx mance metric in biosensing literature, which shows the While designing the MC-Rx, its integrability to a mobile minimum molecular concentration in the vicinity of the nanomachine with limited computational, memory and energy biosensor needed for distinguishing between the existence resources, that requires to operate independently in an MC and the absence of target molecules [156]. Since the input setup, must be taken into consideration. This dictates the fol- signal in MC-Rxs is a physical probability of information lowing requirements for the functionality and physical design carriers, LoD corresponds to the sensitivity metric used of the receiver [17]. for evaluating the performance of EM communication • In situ operation: In-device processing of the molec- receivers, which indicates the minimum input signal ular message must be one of the specifications of the power needed to generate a specified SNR at the output receiver since it cannot rely on any post-processing of of the device. the transduced signals by an external macroscale device • Selectivity: This metric is defined based on the relative or a human controller. affinity of the biorecognition unit to the information • Label-free detection: Detection of information carrying molecules and interferer molecules [157], which can molecules, i.e., IMs, must be done based on their intrin- be non-IMs in the medium or other type of IMs in sic characteristics, i.e., no additional molecular labeling case of MC system with multiple IMs such as MoSK. procedure or preparation stage must be required. High selectivity, i.e., very lower probability of interferer- • Continuous operation: MC-Rx requires to observe the receptor binding compared to target molecule-receptor molecular channel continuously to detect the signal en- binding, is needed to uniquely detect the information coded into concentration, type/ratio/order or release time molecules in the vicinity of the MC-Rx. of molecules. Thus, the functionality of molecular an- • Operation range: The biorecognition unit does not pro- tenna and the processing unit should not be interrupted. vide infinite range of molecular concentration detection Since receptors are needed in the biorecognition unit for as it has finite receptor density. The response of the device sensing target molecules, it is important to have re-usable can be divided into two regions, (i) the linear operation, 16

Intracellular Receiver which is the range of molecular concentration in the Molecular Channel vicinity of the MC-Rx that does not lead to saturation Environment RNAP of receptors, and (ii) the saturation region [158]. When Activator Receptor the information is encoded into the concentration of Complex molecules, the receiver must work in the linear operation region as the output of molecular antenna provides better

information about changes in the molecular concentra- 푃표푃푆표푢푡 tion. However, for other encoding mechanisms, e.g., 푃푅푥

MoSK, the receiver can work in both regions. Activator • Molecular sensitivity: In addition to the aforementioned 푃표푃푆푎푢푥 Cite sensitivity metric that is mapped to LoD for MC-Rxs, Information Receptor a molecular sensitivity can also be defined for an MC- Molecules Activator Rx. This metric indicates the smallest difference in the Expression concentration of molecules that can be detected by an (a) Biological circuit in intracellular environment of an MC-Rx MC-Rx [157]. It is of utmost importance when the

information is encoded in molecular concentration. The Output ሾ푆ሿ푅푥 Ligand-receptor ሾ푅푆ሿ 푃표푃푆표푢푡 metric can be defined as the ratio of changes in the Transcription Binding output of the molecular receiver antenna to changes in Activation the molecular concentration in the vicinity of the MC-Rx ሾ푅ሿ Receptor when the device performs in the linear operation region. 푃표푃푆푎푢푥 Activator ሾ푃푅푥ሿ ሾ푅푁퐴푃ሿ • Temporal resolution: This metric is defined to evaluate Expression the speed of sampling the molecular concentration by the receiver. Since the electrical processes are much faster (b) Functional block diagram than molecular processes, it is expected that the diffusion and binding kinetics limit the temporal resolution. Thus, Fig. 7: Biological circuit of an MC-Rx [168]. biorecognition unit should be realized in transport-limited manner to detect all the messages carried by molecules into the vicinity of the MC-Rx, i.e., the binding kinetics computations, in synthetic cells [18]. Synthetic gene networks should not be a limiting factor on the sampling rate [17]. integrating computation and memory is also proven feasible [19]. More importantly in this context, the technology en- ables implementing bio-nanomachines capable of observing C. Physical Design of MC Receiver individual receptors, as naturally done by living cells; thus, Most of existing studies on the performance of MC ignore stands as a suitable domain for practically implementing more the physical design of the receiver and assume that the receiver information-efficient MC detectors based on the binding state can perfectly count the number of molecules that (i) enter history of individual receptors, as discussed in Section III-D. a reception space with transparent boundaries [11], [159]– In DNAs, gene expression is the process that produces a [161], (ii) hit a 3-D sphere that absorbs molecules [162]–[164], functional gene product, such as a protein. The rate of this or (iii) bind to receptors located on its surface [165]–[167]. expression can be controlled by binding of another protein However, the processes involved in the molecular-to-electrical to the regulatory sequence of the gene. In biological circuits, transduction affect the performance of the receiver, thus, activation and repression mechanisms, that regulate the gene the comprehensive communication theoretical modeling of expressions, are used to connect DNA genes together, i.e., the these processes are required. Available approaches in physical gene expression of a DNA generates a protein which can then design of MC-Rxs can be categorized into two main groups bind to the regulatory sequence of next DNA to control its as follows. expression [169]. In [170], polymerases per second (PoPS) is 1) Biological MC-Rx Architectures: Synthetic biology, the defined as the unit of input and output of a biological cell, engineering of biological networks inside living cells, has i.e., the circuit processes a PoPS signal as input and generates seen remarkable advancements in the last decade, such that another PoPS signal at its output using the aforementioned it becomes possible to device engineered cells, e.g., bacteria, connections among DNAs. The literature in applying synthetic for use as biological machines, e.g., sensors and actuators, biology tools to design bacteria-based MC-Rxs is scarce. An for various applications. Synthetic biology also stands as a MC biotransceiver architecture integrating molecular sensing, promising means of devising nanoscale biotransceivers for transmitting, receiving and processing functions through ge- IoNT applications, by implementing transmission and re- netic circuits is introduced in [31]. However, the analysis is ception functionalities within living cells by modifying the based on the assumption of linearity and time-invariance of the natural gene circuits or creating new synthetic ones [31]. gene translation networks, and does not provide any insight The technology is already mature enough to allow performing into the associated noise sources. The necessary biological complex digital computations, e.g., with networks of genetic elements for an MC-Rx are shown in Fig. 7 [168]. The process NAND and NOR gates, as well as analog computations, such of receiving information is initiated by Receptor Activation as logarithmically linear addition, ratiometric and power-law Expression, which gets a PoPS auxiliary signal as its input 17 and generates receptor proteins, denoted by R, at its output. Among existing biosensing options, the electrical biosen- The incoming signal molecules from molecular channel, i.e., sors are mainly under the focus of MC-Rx design [17]. SRx, then bind to these receptors in Ligand-Receptor Binding The remaining options, i.e., optical and mechanical sensing, unit and form activator complexes, called RS. Finally, the need macroscale excitation and detection units [181]–[183] concentration of RS initiates the Output Transcription Acti- making them inappropriate for an MC-Rx that requires in situ vation, in which RNA polymerase proteins, RNAP , bind to operation. Biocatalytic [184] and affinity-based [185] sensors to the promoter sequence, PRx and produce the PoPS output are two types of electrical biosensors with different molecular signal, P oP SOut. Assuming that the intracellular receiver recognition methods. Biocatalytic recognition is based on two environment is chemically homogeneous, the transfer function steps. First, an enzyme, immobilized on the device, binds of aforementioned biological circuit is derived in [168] to with the target molecule producing an electroactive specie, provide a system theoretic model for the MC-Rx. Genetic such as hydrogen ion. Arrival of this specie near the working circuits are shown to provide higher efficiency for analog electrode of the transducer is then being sensed as it modulates computation compared to digital computations [171]. Thus, one of electrical characteristics of the device. Glucose and analog computing functionalities of genetic circuits are utilized glutamate sensors are examples of the biocatalytic electrical in [154] to derive an analog parity-check decoder circuit. sensors [184], [186]. Alternatively, binding of receptor-ligand The major challenge in use of genetic circuits for imple- pairs on the recognition layer of the sensor is the foundation menting MC-Rx arises from the fact that information transmis- of affinity-based sensing [185]. sion in biological cells is through molecules and biochemical Affinity-based sensing, which is feasible for a a wider reactions. This results in nonlinear input-output behaviors with range of target molecules, such as receptor proteins, and ap- system-evolution-dependent stochastic effects that are needed tamer/DNAs [185], [187], provides a less complicated sensing to be comprehensively studied to evaluate the performance of scheme compared to biocatalytic-based sensing. For example, the device. In [172]–[174], initiative studies on mathematical in the biocatalytic-based sensing, the impact of the additional modeling of cellular signaling are provided. However, it is products of the reaction between target molecule and enzyme shown in [175] that the characterization of the communication in biocatalytic-based sensing on the performance of the device performance of these systems is not analytically tractable. and the application environment must be analyzed thoroughly. Additionally, a computational approach is proposed to charac- Hence, affinity-based recognition is more appropriate for the terize the information exchange in a bio-circuit-based receiver design of a general MC system. However, this recognition using the experimental data published in [176]. The results method is not possible for non-electroactive information- in [175] reveal that the rate of information transfer through carrying molecules, such as glutamate and acetylcholine, biocircuit based systems is extremely limited by the existing which are highly important neurotransmitters in the mam- noises in these systems. Thus, comprehensive studies are malian central nervous system [188]. Thus, it is essential to needed to characterize these noises and derive methods to study the design of biocatalytic-based electrical biosensors as mitigate them. Additionally, existing studies have only focused MC-Rx when these types of information carriers are dictated on the single transmitter-receiver systems, thus, the connection by the application, e.g., communication with neurons. among biological elements for MC with multiple receiver cells Recent advances in nanotechnology led to the design of field remains as an open issue. effect transistor (FET)-based biosensors (bioFETs) providing 2) Nanomaterial-based Artificial MC-Rx Architectures: both affinity-based and biocatalytic-based electrical sensing Similar to MC-Rxs, biosensors are also designed to detect with use of nanowires, nanotubes, organic polymers, and the concentration of an analyte in a solution [177]. Thus, graphene as the transducer unit [186], [189]–[195]. Detection existing literature on MC-Rx design are focused on analyzing of target molecules by bioFETs is based on the modulation the suitability and performance of available biosensing options of transducer conductivity as a result of either affinity-based for receiving the information in MC paradigm [17], [83], or biocatalytic-based sensing. Simple operation principles [178]–[180]. In this direction, researchers must consider the together with the extensive literature on FETs established fundamental differences between a biosensor and a MC-Rx, through many years, electrical controllability of the main arising from their different application areas, as stated below. device parameters, high-level integrability, and plethora of • Biosensors are designed to perform typically in equilib- optimization options for varying applications make FET-based rium condition. However, MC-Rxs must continuously ob- biosensing technology also a quite promising approach for serve the environment and detect the information encoded electrical MC-Rx. Moreover, these sensors promise label-free, into a physical property related to the molecules, such as continuous and in situ operation in nanoscale dimensions. concentration, type/ratio/order, or arrival time. Thus, the design of an MC-Rx based on the principles of • Biosensors are mostly designed for laboratory applica- affinity-based bioFETs is the main approach considered in the tions with macroscale readout devices and human ob- literature [17], [178], which will be overviewed in the rest of servers to compensate the lack of an integrated processor, this section. which is not applicable for an MC-Rx. a) Nanoscale bioFET-based MC-Rx Architectures: As Thus, while the biosensing literature provides insights for MC- shown in Fig. 8, a bioFET consists of source and drain Rx design, ICT requirements of the device and its performance electrodes and a transducer channel, which is functionalized in an MC paradigm must be considered to reach appropriate by ligand receptors in affinity-based sensors [187]. In this solutions. type of sensors, the binding of target molecules or analytes 18

option to control the selectivity of the MC-Rx. The appropriate Received aptamer for a target ligand can be find using SELEX process, Insulator information i.e., searching a large library of DNAs and RNAs to determine molecules a convenient nucleic acid sequence [202]. Material used for transducer channel: One of the most Receptors important bioFET design parameters is the material used as the transducer channel between source and drain electrodes, which determines the receiver geometry and affects the electrical noise characteristics of the device. Nanowires (NWs) [193], Source [203], single walled carbon nanotubes (SWCNTs), graphene electrode [204], molybdenum disulfide (MoS2) [205], and organic ma- terials like conducting polymers [206] are some examples SiO Transducer 2 of nanomaterials suitable for use in a bioFET channel. In Si channel Drain the first generation of bioFETs, one dimensional materials electrode such as SWCNT and NW were used as the channel in a bulk form. However, use in the form of single material or aligned arrays outperformed the bulk channels in terms of Fig. 8: Conceptual design of a bioFET-based MC-Rx. sensitivity and reduced noise [207]. Among possible NW materials, such as SnO2, ZnO and In2O3 [208], silicon NW (SiNW) bioFETs have shown high sensitivity, high integration to the ligand receptors leads to accumulation or depletion of density, high speed sampling and low power consumption the carriers on the semiconductor channel. This modulates [209]–[214]. However, their reliable and cost-effective fab- the transducer conductivity, which, in turn, alters the flow rication is still an important open challenge [190], [215]. of current between source and drain. Thus, by fixing the Comprehensive reviews exist on the performance of SiNW source-to-drain potential, the current flow becomes a function bioFETs, their functionality in biomedical applications such of the analyte density and the amount of analyte charges. as disease diagnostic, their top-down and bottom-up fabrica- Since the ligand receptors are being selected according to tion paradigms and integration within complementary metal- the target molecule, i.e., ligands, bioFETs do not require any oxide-semiconductor (CMOS) technology [207], [214], [216]– complicated post-processing such as labeling of molecules. [218]. It is concluded that SiNW bioFETs must be designed Moreover, the measurement of source-drain current does not according to the requirements arose by applications since need any macroscale readout unit. Thus, bioFETs can be used defining an ideal characteristic for these devices is difficult. for direct, label-free, continuous and in situ sensing of the Additionally, both theoretical and experimental studies are molecules in an environment as a standalone device. needed to find the impact of structure parameters on their One of significant advantages of bioFETs over other elec- functionality. Moreover, fine balancing of important structural trical sensors is their wide range of design parameters. A factors, such as number of NWs, their doping concentration list of FET-based biosensors and their applications in sensing and length, in SiNW bioFETs design and fabrication remains a different type of molecules is provided in Table III. In the challenge, which affects the sensitivity, reliability and stability following, we further describe these vast design options. of the device. SWCNT-based bioFETs offer higher detection Type of bioreceptors: First important design parameter sensitivity due to their electrical characteristic, however, these arises from the type of receptors used in the Biorecognition devices also face fabrication challenges, such that their defect Unit, which causes the selectivity of receiver for a certain type free fabrication is the most challenging among all candidates of molecules that will be used as information carrier in MC [219]. Note that existence of defects can adversely affect paradigm. Among possible receptor types for affinity-based the performance of SWCNT bioFETs in an MC-Rx. More- bioFETs, natural receptor proteins and aptamer/DNAs are ap- over, for in vivo applications, the biocompatibility of CNTs propriate ones for an MC-Rx, since their binding to the target and biodurability of functionalized SWCNTs are still under molecule is reversible and their size is small enough to be doubt [220]–[222]. While both NWs and CNTs have one- used in a nanomachine [185], [187]. As an example, the FET dimensional structure, use of two-dimensional materials as the transducer channel is functionalized with natural receptors, transducer channel leads to higher sensitivity, since a planar e.g., neuroreceptors, to detect taste in bioelectronic tongues structure provides higher spatial coverage, more bioreceptors [199] and odorant in olfactory biosensors [200]. An advantage can be functionalized to its surface and all of its surface of these type of receptors is their biocompatibility that makes atoms can closely interact with the bond molecules. Thus, them suitable for in vivo applications. In addition, use of graphene, with its extraordinary electrical, mechanical and aptamers, i.e., artificial single-stranded DNAs and RNAs, in chemical characteristics, is a promising alternative for the the recognition unit of bioFETs provides detectors for a wide transducer channel of bioFETs [215], [223]. range of targets, such as small molecules, proteins, ions, Intrinsic flexibility of graphene provides higher chance aminoacids and other oligonucleotides [190], [198], [201]. of integration into devices with non-planar surfaces which Since an immense number of aptamer-ligand combinations can be more suitable for the design of nanomachines in an with different affinities exist, it provides a powerful design MC application [204]. There is currently tremendous amount 19

TABLE III: Design Options, Performance and Applications of bioFETs

Transducer Limit of Bioreceptor Application Ref. channel detection SWCNT Acetylcholine(Ach) receptor 100 pM ACh detection [195] SWCNT Receptor protein (antibody) 1 ng/ml Prostate cancer detection [194] SWCNT Anti Carcinoembryonic antigen (CEA) ∼ 55 pM CEA detection [196] SiNW Receptor protein (antibody) ∼2 fM Prostate cancer detection [193] SiNW Estrogen receptors 10 fM dsDNA detection [197] ZnO NW Anti-Immunoglobulin G(IgG) antibodies ∼0.3 nM IgG antibodies sensing [192] Graphene Glucose oxidase (enzyme) 0.1 mM Glucose sensor [186] Graphene Glutamic dehydrogenase(enzyme) 5 µM Glutamate sensor [186] Graphene Pyrene-linked peptide nucleic acid (pPNA) 2 pM DNA sensor [191] Graphene Single-stranded probe DNA 10 pM DNA sensor [190] Graphene Immunoglobulin E (IgE) aptamers 0.3 nM IgE protein detection [198] MoS2 Glucose oxidase (enzyme) 300 nM Glucose sensor [189]

Background and Transducing Noise noise source in low frequencies as it increases with decreasing Binding Noise Molecular Output the frequency [17]. Detailed information on the impact of Signal Biorecognition Transducer Output Signal the aforementioned noise processes on the performance of Unit Unit Unit bioFETs can be found in [233], [234]. Moreover, experimental Fig. 9: Block diagram of a bioFET-based MC-Rx. studies are provided in [235] on the noise resulted from ion dynamic processes related to ligand-receptor binding events of a liquid-gated SiNW array FET by measuring the noise spectra of interest in building different configurations of graphene of device before and after binding of target molecules. bioFETs, e.g., back-gated [224] or solution-gated [223], with While the existing biosensing literature can provide insight research showing its superior sensing performance for various for the MC-Rx design, there is a need for investigation of analytes, e.g., antigens [225], DNA [226], bacteria [227], design options according to the communication theoretical odorant compounds [223], and glucose [228]. requirements of an MC-Rx. Few studies have focused on General block diagram of a bioFET-based MC-Rx is shown evaluating the performance of bioFETs in an MC paradigm. in Fig. 9 . The biorecognition Unit is the interface between A SiNW bioFET-based MC-Rx is modeled in [17] based on communication channel and the receiver, thus, it models equilibrium assumption for the receptor-ligand reaction at the sensing of the concentration of ligands. The random motion receiver surface. The study provides a circuit model for the of ligands near the surface of the receiver, which is governed transducer unit of the receiver. This work is further extended by Brownian motion, leads to fluctuations in the number of in [178], where the spatial and temporal correlation effects bound receptors. This fluctuation can be modeled as a binding resulting from finite-rate transport of ligands to the stochastic noise, which depends on the transmitted signal and can ad- ligand-receptor binding process are considered to derive the versely affect the detection of the ligands concentration [229]. receiver model and its noise statistics. In [236], an MC-Rx The communication theoretical models of binding noise are consisting of an aerosol sampler, a SiNW bioFET function- presented in [165], [230]. Background noise, also called bio- alized with antibodies and a detection stage is designed for logical interference [17], is resulted from binding of molecules virus detection. The performance of the receiver is studied different from targeted ligands that might exist in the com- by considering the system in steady state. While the receiver munication channel and show similar affinity for the receptors model in [236] takes into account the flicker noise and thermal [229]. Note that this is different from intersymbol interference noise, it neglects the interference noise by assuming that the and co-channel interference studied in [231], [232]. Stochastic MC-Rx performs in a perfectly sanitized room. Moreover, binding of ligands to the receptors modulated the conductance the models used in all of these studies assume the ligand- of the FET channel, which is modeled by Transducer unit in receptor binding process in thermal equilibrium and they do Fig. 9. This conductance changes is then reflected into the not capture well the correlations resulting from the time- current flowing between source and drain electrodes of FET varying ligand concentration occurring in the case of molecular by Output Unit. The surface potential of the FET channel, thus communications. More importantly, these studies only cover its source-to-drain current, can be affected by undesirable ionic SiNW bioFET-receivers, and do not provide much insight adsorptions in application with ionic solutions [17]. Hence, a into the performance of other nanomaterials as the transducer reference electrode can be used in the solution to stabilize the channel, such as graphene that promises to provide higher surface potential [157]. Finally, the transducing noise shown detection sensitivity due to its two-dimensional structure. in Fig. 9 covers the impact of noise added to the received Thus, the literature misses stochastic models for signal during transducing operation including thermal noise, nanomaterial-based MC nanoreceiver architectures, which caused by thermal fluctuations of charge carriers on the bound are needed to study the performance of the receiver in MC ligands, and 1/f (flicker) noise, resulted from traps and defects scenarios, i.e., when the device is exposed to time-varying in the FET channel [233]. Flicker noise can be the dominant concentration signals of different types and amplitudes. 20

These models must capture the impacts of receiver geometry, i.e., length [64], dumbbell hairpins [61], [65], and short its operation voltage characteristics and all fundamental sequence motifs/labels [62]. For information detection, solid- processes involved in sensing of molecular concentration, state [62] and DNA-origami [68] based nanopores can be such as molecular transport, ligand-receptor binding kinetics, utilized to distinguish information symbols, i.e., the properties molecular-to-electrical transduction by changes in the of DNA/RNA strands by examining the current characteristics conductance of the channel. To provide such a model for while DNA/RNA strands pass through the nanopores. As pre- graphene-based bioFETs, the major factors that influence sented in Fig. 3, MC-Rx contains receptor nanopores through the graphene properties must be taken into account. The which these negatively charged DNA strands pass thanks to number of layers is the most dominant factor since electronic the applied potential, and as they do, they obstruct ionic cur- band structure, which has a direct impact on the electrical rents that normally flow through. The duration of the current properties of the device, is more complex for graphene with obstruction is proportional to the length of the DNA strand that more number of layers [237]–[241]. Next important parameter passes through, which is utilized for selective sensing. The use is the substrate used in graphene-based bioFET, specially of nanopores for DNA/RNA symbol detection also enables the when the number of layers is less than three [241], [242]. The miniaturization of MC capable devices towards the realization carrier mobility in graphene sheet is reported to be reduced of IoNT. According to [61], 3-bit barcode-coded DNA strands by more than an order of magnitude on SiO2 substrate due with dumbbell hairpins can be detected through nanopores to charged impurities in the substrate and remote interfacial with 94% accuracy. In [66], four different RNA molecules phonon scattering [243]. On the other hand, it is shown that having different orientations are translocated with more than impacts of substrate can be reduced in suspended single-layer 90% accuracy while passing through transmembrane protein graphene sheet, resulting in higher carrier mobility [244]. nanopores. The time of the translocation event depends on the Moreover, as a result of graphene’s large surface area, the voltage, concentration and length of the DNA symbols, and the impact of impurities on its performance can be substantial translocation of symbols can take up to a few ms up to hundred [242]. The atomic type, amounts and functional groups on the ms time frames [61], [65]. Considering the slow diffusion edges of graphene, which are hard to measure and control, channel in MC, transmission/detection of DNA/RNA-encoded are also among the properties that can result in trial to symbols do not introduce a bottleneck and multiple detections trial variations in the performance of the fabricated device can be performed during each symbol transmission. [245]–[247]. Additionally, inherent rippling in graphene sheet, defects and size of sheet also affect the properties of the device [242], [248]–[255]. D. Detection Methods for MC b) Other MC-Rx Architectures: Few studies exist on the practical MC systems taking into account the physical design Detection is one of the fundamental aspects of communica- of the receiver. In [83], the isopropyl alcohol (rubbing alcohol) tions having tremendous impact over the overall communica- is used as information carrier and commercially available tion performance. The detection of MC signals is particularly metal oxide semiconductor alcohol sensors are used as MC- interesting due to the peculiarities of the MC channel and com- Rx. This study provides a test-bed for MC with macroscale municating nanomachines, which impose severe constraints dimensions, which is later on utilized in [256] to estimate its on the design of detection methods. For example, the limited combined channel and receiver model. This test-bed is ex- energy budget and computational capabilities of nanomachines tended to a molecular multiple-input multiple-output (MIMO) due to their physical design restrict the complexity of the system in [257] to improve the achievable data rate. In [179], methods. The memory of the diffusion channel causes severe the information is encoded in pH level of the transmitted fluid ISI and leads to time-varying channel characteristics with and a pH probe sensor is used as the MC-Rx. Since use of very short coherence time. The stochastic nature of Brownian acids and bases for information transmission can adversely motion and sampling of discrete message carriers bring about affect the other processes in the application environment, such different types of noise, e.g., counting noise and receptor as in the body, magnetic nanoparticles (MNs) are used as binding noise. The physiological conditions, in which most of information-carrying molecules in microfluidic channels in the nanonetwork applications are envisioned to operate, imply [180]. In this study, a bulky susceptometer is used to detect the the abundance of molecules with similar characteristics that concentration of MNs and decode the transmitted messages. can lead to strong molecular interference. These challenges In addition, performance of MN-based MC where an external have been addressed in MC to different extents. In this magnetic field is employed to attract the MNs to a passive section, we will provide an overview of the state-of-the-art receiver is analyzed in [258]. MC detection approaches, along with a discussion on their However, the focus of the aforementioned studies is using performances and weaknesses. macroscale and commercially available sensors as receiver. We classify the existing approaches according to the con- Thus, these studies do not contribute to the design of a sidered channel and received signal models, which reflect nanoscale MC-Rx. As discussed in Section II-B1, recent the envisioned device architectures that impose different con- advancements in DNA/RNA sequencing and synthesis tech- straints or allow different simplifications over the problem. niques have enabled DNA/RNA-encoded MC [61], [62]. For Accordingly, we divide the detection methods into two main information transmission, communication symbols can be re- categories: MC detection with passive and absorbing receivers, alized with DNA/RNA strands having different properties, and MC detection with reactive receivers. 21

1) MC Detection with Passive and Absorbing Receivers: The nonlinearities and complexity of the MC system often lead researchers to use simplifying assumptions to develop Passive detection methods and analyze their performances. To this end, Receiver the intricate relationship between molecular propagation and sampling processes is often neglected. Information Transparent Passive receiver (PA) concept is the most widely used sim- molecules boundary plifying assumption in MC literature, as it takes the physical sampling process out of the equation, such that researchers x can focus only on the transport of molecular messages to Absorbing the receiver location. Accordingly, the passive receiver is x Receiver often assumed to be a spherical entity whose membrane is x transparent to all kind of molecules, and it is a perfect observer Absorbed & removed Absorbing of the number of molecules within its spherical reception molecules boundary space, as shown in Fig. 10 [259]. In the passive receiver approximation, the receiver has no impact on the propagation of molecules in the channel. Passive receivers can also be Reactive considered as if they include ligand receptors, which are Receiver homogeneously distributed within the reception space with very high concentration and infinitely high rate of binding with ligands, such that every single molecule in the reception space is effectively bound to a receptor at the time of sampling. Intracellular Receptors Another modeling approach, i.e., absorbing receiver (AB) molecules concept [260], considers receiver as an hypothetical entity, Fig. 10: Hypothetical MC-Rx models used for developing often spherical, which absorbs and degrades every single detection methods. molecule that hits its surface, as demonstrated in Fig. 10. This approach improves the assumption of passive receiver one step further towards a more realistic scenario including a physical geometry is often neglected and assumed to be unbounded, interaction between the receiver and the channel. In contrast and molecules are assumed to propagate independently from to passive receiver, absorbing receiver can be considered to each other. In some studies addressing passive receivers, have receptors located over the surface. For a perfect absorbing researchers consider the existence of enzymes in the channel, receiver, this means very high concentration of receptors with which reduce the impact of the ISI by degrading the residual infinitely high absorption rate, such that every molecule that messenger molecules through first-order reaction [261]. For a hits the surface is bound and consumed instantly. three dimensional free diffusion channel with uniform flow in Physical correspondence of both models is highly question- the presence of degrading enzymes, the number of molecules able. Nevertheless, they are widely utilized in the literature as observed in the spherical reception space of a passive receiver they provide upper performance limits. However, ignoring the follows non-stationary Poisson process [11], [262], i.e., receptor-ligand reactions, which often leads to further intrica- N (t) ∼ Poisson (λ (t)) , (4) cies, e.g., receptor saturation, stands as a major drawback of RX|PA RX these approaches. where the time-varying mean of this process λRX(t) can be a) Received Signal Models: When constructing the re- given by ceived signal models for diffusion-based MC, the transmitter t b T +1c (Tx) geometry is usually neglected assuming that the Tx Xs   is a point source that does not occupy any space. This λRX(t) = λnoise + Q s[i]Pobs t − (j − i)Ts . (5) assumption is deemed valid when the distance between the j=1 Tx and Rx are considerably larger than the physical sizes of The mean depends on the number of transmitted molecules the devices. Throughout this section, we will mostly focus on Q to represent bit 1, the symbols transmitted in the current OOK modulation, where the Tx performs an impulsive release symbol interval as well as in the previous symbol intervals, of a number of molecules to transmit bit-1, and does not send i.e., s[i], and the length of a symbol interval Ts. Most MC any molecule to transmit bit-0. This is the most widely used studies include an additive stationary noise in their models modulation scheme in MC detection studies, as it simplifies representing the interfering molecules available in the channel the problem while capturing the properties of the MC channel. as a result of an independent process in the application envi- However, we will also briefly review the detection schemes ronment. These molecules are assumed to be of the same kind corresponding to other modulation methods, e.g., timing-based with the messenger molecules and their number is represented modulation and MoSK, throughout the section. by a Poisson process and captured by λnoise. Channel response Molecular propagation in the channel is usually assumed to is integrated into the model through the function Pobs(t), which be only through free diffusion, or through the combination of is the probability of a molecule transmitted at time t = 0 to diffusion and uniform flow (or drift). In both cases, the channel be within the sampling space at time t. When Tx-Rx distance 22 is considerably large, ligands are typically assumed to be b) Detection Methods: Detection methods for MC in uniformly distributed within the reception space. As a result, general can be divided into two main categories depending the channel response can be written as on the method of concentration measurement: sampling-based  2  and energy-based detection. Passive receivers are usually as- VRX |~reff| Pobs(t) = exp −kCEt − , (6) sumed to perform sampling-based detection, which is based (4πDt)3/2 4Dt on sampling the instantaneous number of molecules inside 4 3 where VRX = 3 πdRX is the volume of the spherical receiver the reception space at a specific sampling time [270]. Ab- with radius dRX, D is the diffusion coefficient, CE is the uni- sorbing receivers, on the other hand, are typically assumed to form concentration of the degrading enzymes in the channel, utilize energy-based detection, which uses the total number k is the rate of enzymatic reaction, and ~reff is the effective of molecules absorbed by the receiver during a prespecified Tx-Rx distance vector, which captures the effect of uniform time interval, that is usually the symbol interval [269]. In flow [262]. Assuming that the Tx and Rx are located at some studies, passive receivers are also considered to perform ~rTX = (0, 0, 0), ~rRX = (x0, 0, 0), respectively, and the flow energy-based detection through taking multiple independent velocity is given by vx, vy, vz in 3D Cartesian coordinates, samples of number of molecules inside the reception space the magnitude of the effective distance vector can be written at different time instants during a single symbol interval, as follows and passing them through a linear filter which outputs their q 2 + 2 weighted sum as the energy of the received molecular signal |~reff| = (x0 − vxt) + (vy) (vz) . (7) [160], [262]. For an absorbing receiver, the received signal is usually As in conventional wireless communications, detection taken as the number of molecules absorbed by the Rx within can be done on symbol-by-symbol or sequential basis. The a time interval [260]. For a diffusion channel without flow, symbol-by-symbol detection tends to be more practical in the probability density for a molecule emitted at t = 0 to be terms of complexity, whereas the sequence detectors require absorbed by a perfectly absorbing receiver of radius rr and the receiver to have a memory to store the previously de- located at a distance r from the Tx at time t is given by coded symbols. Due to the MC channel memory causing a  2  rr 1 r − rr (r − rr) considerable amount of ISI for high data rate communication, fhit(t) = √ exp − , (8) r 4πDt t 4Dt the sequence detectors are more frequently studied in the literature. and the cumulative distribution function (CDF) is given by Next, we review the existing MC detection techniques Z t   0 0 rr r − rr developed for passive and absorbing receivers by categorizing Fhit(t) = fhit(t )dt = erfc √ , (9) 0 r 4Dt them into different areas depending on their most salient where erfc is the complementary error function [260]. The characteristics. A comparison matrix for these methods can CDF can be used to calculate the probability of a molecule also be seen in Table IV. transmitted at time t = 0 to be absorbed within the kth Symbol-by-Symbol (SbS) Detection: SbS MC detectors in signaling interval, i.e., the literature are usually proposed for very low-rate commu- nication scenarios, where the ISI can be neglected, asymptoti- Pk = Fhit(kTs) − Fhit([k − 1]Ts). (10) cally included into the received signal model with a stationary When considering multiple independent molecules emitted at mean and variance, or approximated by the weighted sum the same time, the number of molecules absorbed at the kth of ISI contributions of a few previously transmitted sym- interval becomes Bernoulli random variable with the success bols. In [159], a one-shot detector is proposed based on the probability of Pk. Assuming that the success probability is asymptotic approximation of the ISI assuming that the sum low enough, Gaussian approximation of the Bernoulli random of decreasing ISI contributions of the previously transmitted process can be used to write symbols can be represented by a Gaussian distribution through central limit theorem (CLT) based on Lindeberg’s condition. N [i] ∼ N µ[i], σ2[i]
, (11) RX|AB A fixed-threshold detector is proposed maximizing the mu- where its signal-dependent mean and variance can be written tual information between transmitted and decoded symbols. as a function of current and previously transmitted bits s[i], Similarly, in [160], [262], a matched filter in the form of a i.e., weighted sum detector is proposed using a different asymptotic k ISI approximation as though it results from a continuously X µ[i] = Q Pks[i − k + 1], (12) emitting source leading to a stationary Poisson distribution i of interference molecules inside the reception space. In this k scheme, a passive receiver performs energy-based detection 2 2 X σ [i] = σnoise + Q Pk(1 − Pk)s[i − k + 1]. (13) taking multiple samples at equally spaced sampling times i during a single symbol transmission, and the weights of the Note that as in the case of passive receiver, the received signal samples are adjusted according to the number of molecules model includes the contribution of a stationary noise through expected at the corresponding sampling times. This matched 2 its variance σnoise. Unfortunately, in the literature, there is no filter is proven to be optimal in the sense that it maximizes analytical model for absorbing receivers in diffusion-based SNR at the receiver. However, the optimal threshold of this MC channels with uniform flow and degrading enzymes. detector does not lend itself to a closed-form expression, 23

TABLE IV: Comparison Matrix for MC Detectors with Passive and Absorbing Receivers

Channel Modulation & Detector Rx Measurement CSI Description Complex. Perf. Characteristics Transmit Waveform Type Type Method Req.

One-shot fixed-threshold detector [159] Diffusion CSK/OOK-Impulse SbS PA Sampling Ins. CSI Low Sub. Weighted sum detector [263] Diff./Flow/En./EI CSK/OOK-Impulse SbS PA Sampling Ins. CSI Low Opt. Weighted sum detector [160], [262] Diff./Flow/En./EI CSK/OOK-Impulse SbS PA Sampling Ins. CSI Low Sub. Adaptive-threshold detector [264] Diff./EI CSK/OOK-Impulse SbS PA Sampling No CSI Moderate Sub. Adaptive-threshold detector [265]–[267] Diffusion CSK/OOK-Impulse SbS AB Energy No CSI Low Sub. Linear equalizer (MMSE) [11] Diffusion CSK/OOK-Pulse Seq. PA Sampling Ins. CSI Very High Sub. Nonlinear equalizer (DFE) [11] Diffusion CSK/OOK-Pulse Seq. PA Sampling Ins. CSI Very High Sub. ML sequence detector with Viterbi [11] Diffusion CSK/OOK-Pulse Seq. PA Sampling Ins. CSI Very High Opt. ML sequence detector with Viterbi [160] Diff./Flow/En./EI CSK/OOK-Impulse Seq. PA Sampling Ins. CSI Very High Opt. ML sequence detector with Viterbi [268] Diffusion MoSK-Impulse Seq. PA Sampling Ins. CSI Very High Opt. Near ML sequence det. with RS Viterbi [159] Diffusion CSK/OOK-Impulse Seq. PA Sampling Ins. CSI High Sub. Strength-based ML detector [269] Diffusion CSK/ASK-Impulse SbS PA Energy Ins. CSI High Opt. Sampling-based ML detector [270] Diffusion CSK/ASK-Impulse SbS PA Sampling Ins. CSI High Opt. Derivative-based detector [271] Diffusion CSK/OOK-Impulse SbS AB Energy Ins. CSI Moderate Sub. Noncoherent detector [272] Diffusion CSK/B-CSK-Pulse SbS PA Sampling No CSI Low Sub. Local convexity-based noncoherent det. [273] Diffusion CSK/OOK-Pulse SbS PA Sampling No CSI Moderate Sub. Noncoherent ML threshold-based det. [274] Diff./EI CSK/OOK-Impulse SbS PA&AB Sampling Stat. CSI Low Opt. Noncoherent ML sequence detector [274] Diff./EI CSK/OOK-Impulse Seq. PA&AB Sampling Stat. CSI High Opt. Noncoherent decision-feedback detector [274] Diff./EI CSK/OOK-Impulse Seq. PA&AB Sampling Stat. CSI Very High Sub. Noncoherent blind detector [274] Diff./EI CSK/OOK-Impulse SbS PA&AB Sampling No CSI Low Sub. ML sequence detector with CC codes [275] Diff./EI CSK/ASK-Impulse Seq. PA Sampling No CSI Very High Opt. Asynchronous fixed-threshold detector [276] Diffusion CSK/OOK-Impulse SbS PA Sampling Ins. CSI Moderate Sub. Asynchronous adaptive-th. det. with DF [276] Diffusion CSK/OOK-Impulse SbS PA Sampling Ins. CSI High Sub. Adaptive-threshold detector (Mobile MC) [277] Diff./TV CSK/OOK-Impulse SbS PA Sampling Ins. CSI Very High Sub. Single-sample th. det. (Mobile MC) [278] Diff./Flow/TV CSK/OOK-Impulse SbS PA Sampling Out. CSI High Sub. ML sequence detector (Timing Channel) [279] Diffusion Release time-Impulse Seq. AB Arrival time Ins. CSI Very High Opt. Sequence detector with modified Viterbi [279] Diffusion Release time-Impulse Seq. AB Arrival time Ins. CSI High Sub. Symbol by symbol timing detector [279] Diffusion Release time-Impulse SbS AB Arrival time No CSI Low Sub. ML detector with FA & LA times [123], [280] Diffusion Release time-Impulse SbS AB Arrival time Ins. CSI Moderate Opt. Abbreviations — Diff.: Diffusion – En.: Enzyme – EI: External Interference – TV: Time-varying – SbS: Symbol-by-symbol detector – Seq.: Sequential detector – PA: Passive receiver – AB: Absorbing receiver – Ins. CSI: Instantaneous CSI – Stat. CSI: Statistical CSI – Out. CSI: Outdated CSI – Sub.: Sub-optimal – Opt. Optimal – Complex.: Complexity – Perf.: Performance – DF: Decision-Feedback – FA: First Arrival – LA: Last Arrival – CC codes: Constant Composition codes – RS: Reduced-State and thus, it should be numerically obtained through resource sequence detection methods. Similarly, a near-optimal ML intensive search algorithms. Similarly, in [263], consider- sequence detector employing Viterbi algorithm is proposed in ing also the external sources of interference, another linear [159]. Another optimal ML sequence detector is proposed in matched filter is designed maximizing the expected signal- [160] for MC with uniform flow and enzymes that degrade to-interference-plus-noise-ratio (SINR) for SbS detection, and information molecules. shown to outperform previous schemes especially when the In addition to the sequence detection methods and equal- ISI is severe. There are also adaptive-threshold-based SbS de- izers, there are other approaches proposed to overcome the tection methods relying on receivers with memory of varying effects of the ISI on detection. For example, in [281], authors length taking into account only the ISI contribution of a finite propose to shift the sampling time by increasing the reception number of previously transmitted symbols [264]–[267], [269], delay to reduce the effect of ISI. In [271], a derivative-based [270]. In these schemes, the adaptive threshold is updated for signal detection method is proposed to enable high data rate each symbol interval using the ISI estimation based on the transmission. The method is based on detecting the incoming previously decoded symbols. SbS detection is also considered messages relying on the derivative of the channel impulse in [274], [276], which will be discussed next in the context of response (CIR). noncoherent and asynchronous detection. Noncoherent Detection: Most of the MC detection methods Sequence Detection and ISI Mitigation: Optimal sequence requires the knowledge of the instantaneous CSI in terms detection methods based on Maximum a Posteriori (MAP) and of CIR. However, CIR in MC, especially in physiologically Maximum Likelihood (ML) criteria are proposed in [11] for relevant conditions, tends to change frequently, rendering the MC with passive receivers. Even though the complexity of the detection methods relying on the exact CIR knowledge useless. sequence detectors are reduced by applying Viterbi algorithm, Estimating the instantaneous CIR is difficult and requires high it still grows exponentially with increasing channel memory computational power. To overcome this problem, researchers length. To reduce the complexity further, a sub-optimal linear propose low-complexity noncoherent detection techniques. For equalizer based on Minimum Mean-Square Error (MMSE) example, in [267], the authors develop a simple detection criterion is proposed. To improve the performance of the method for absorbing receivers, which does not require the sub-optimal detection, a nonlinear equalizer, i.e., Decision- channel knowledge. In this scheme, the receiver performs Feedback Equalizer (DFE), is also proposed in the same a threshold-based detection by comparing the number of study. DFE is shown to outperform linear equalizers with absorbed molecules in the current interval to that of the significantly less complexity than optimal ML and MAP previous symbol interval. The adaptive threshold is updated 24 in every step of detection with the number of molecules as well. absorbed. However, this method performs poorly when a Diffusion-based molecular timing (DBMT) channels are sequence of consecutive bit-1s arrives. Similarly, in [272], also addressed from detection theoretical perspective. DBMT the difference of the accumulated concentration between two channels without flow are accompanied by a Levy distributed adjacent time intervals is exploited for noncoherent detection. additive noise having a heavy algebraic tail in contrast to the In [273], the local convexity of the diffusion-based channel exponential tail of inverse Gaussian distribution, which DBMT response is exploited to detect MC signals in a noncoherent channel with flow follows [280]. In [279], an optimal ML manner. A convexity metric is defined as the test statistics, detector is derived for DBMT channels without flow; however, and the corresponding threshold is derived. There are also the complexity of the detector is shown to have exponen- methods requiring only the statistical CSI rather than the tial computational complexity. Therefore, they propose sub- instantaneous CSI [274]. Additionally, constant-composition optimal yet practical SbS and sequence detectors based on codes are proposed to enable ML detection without statistical the random time of arrivals of the simultaneously released or instantaneous CSI, and shown to outperform uncoded information molecules, and show that the performance of transmission with optimal coherent and noncoherent detection, the sequence detector is close to the one of computationally when the ISI is neglected [275]. expensive optimal ML detector. Asynchronous Detection: The synchronization between the In DBMT channels without flow, linear filtering at the re- communicating devices is another major challenge. However, ceiver results in a dispersion larger or equal to the dispersion of in the previously discussed studies, synchronization is assumed the original, i.e., unfiltered, sample, rendering the performance to be perfect. To overcome this limitation, an asynchronous of releasing multiple particles worse than releasing a single peak detection method is developed in [276] for the de- particle. Based on this finding, the authors in [280] develop modulation of MC signals. Two variants has been proposed. a low-complexity detector, which is based on the first arrival First method is based on measuring the largest observation (FA) time of simultaneously released particles by the TX. The within a sampling interval. This SbS detection method is of method is based on the observation that the probability density moderate complexity and non-adaptive, comparing the maxi- of the FA gets concentrated around the transmission time when mum observation to a fixed threshold. The second method is the number of released molecules M increases. Neglecting adaptive and equipped with decision feedback to remove the ISI, it is shown in the same paper that the proposed FA-based ISI contribution. In this scheme, the receiver takes multiple detector performs very close to the optimal ML detectors for samples per bit and adjusts the threshold for each observation small values of M. However, the ML detection still performs based on the expected ISI. significantly better than the FA for high values of M. The Detection for Mobile MC: The majority of MC studies detection based on the order statistics has been extended in assumes that the positions of Tx and Rx are static during the same authors’ later work [123], where they consider also communication. The mobility problem of MC devices has just the detection based on the last arrival (LA) time. Defining a recently started to attract researchers’ attention. For example, system diversity gain as the asymptotic exponential decrease MC between a static transmitter and a mobile receiver is rate of error probability with the increased number of released considered in [277], where the authors propose to reconstruct particles, they showed that the diversity gain of LA detector the CIR in each symbol interval using the time-varying approaches to that of computationally expensive ML detector. transmitter-receiver distance estimated based on the peak value 2) MC Detection with Reactive Receivers: Another type of the sampled concentration. Two adaptive schemes, i.e., of receiver considered for MC is reactive receiver, which concentration-based adaptive threshold detection and peak- samples the molecular concentration of incoming messages time-based adaptive detection, are developed based on the through a set of reactions it performs via specialized receptor reconstructed CIR. In [278], different mobility cases including proteins or enzymes, as shown in Fig. 10. The reactive mobile TX and RX, mobile TX and fixed RX, and mobile receiver approach is more realistic in the sense that natural RX and fixed TX are considered to develop a stochastic cells, e.g., bacteria and neurons, sense molecular communi- channel model for diffusive mobile MC systems. The authors cation signals through their receptors on the cell membrane, derive analytical expressions for the mean, PDF, and auto- and many types of artificial biosensors, e.g., bioFETs, are correlation function (ACF) of the time-varying CIR, through functionalized with biological receptors for higher selectivity. an approximation of the CIR with a log-normal distribution. Since synthetic biology, focusing on using and extending Based on this approximation, a simple model for outdated CSI natural cell functionalities, and artificial biosensing are the two is derived, and the detection performance of a single-sample phenomena that are considered for practical implementation of threshold detector relying on the outdated CSI is evaluated. MC receivers, studying MC detection with reactive receivers Discussion on Other Detection Techniques: MC detection has more physical correspondence. problem is also addressed for molecule shift keying (MoSK) Diffusion-based MC systems with reactive receivers, in most modulation. In [268], an optimal ML sequence detector em- cases, can be considered as reaction-diffusion (RD) systems ploying Viterbi algorithm is proposed assuming that a passive with finite reaction rates. Although RD systems, which are typ- receiver can independently observe MC signals carried by dif- ically highly nonlinear, have been studied in the literature for ferent types of molecules. This assumption greatly simplifies a long time, they do not usually lend themselves to analytical the problem and enable the application of detection methods solutions, especially when the spatio-temporal dynamics and developed for CSK-modulated MC signals for MoSK signals correlations are not negligible. To be able to devise detection 25 methods and evaluate their performance in the MC framework, as long as the collected n samples of unbound and bound time researchers have come up with different modeling approaches, intervals are independent. These observable characteristics of which will be reviewed next. For the sake of brevity, we focus the ligand-receptor binding reactions have been exploited to our review on detection with receivers equipped with ligand infer the incoming messages to different extents, as will be receptors, which have only one binding site. reviewed next. Ligand-receptor binding reaction for a single receptor ex- a) Received Signal Models: The nonlinearities arising posed to time-varying ligand concentration cL(t) can be from the interaction of time-varying MC signals with receptors schematically demonstrated as follows have led to different approaches for modeling MC systems with reactive receivers compromising on different aspects to cL(t)k+ U )−−−−−−−−−−*B, (14) develop detection techniques and make the performance anal- k − yses tractable. A brief review of these modeling approaches where k+ and k− are the ligand-receptor binding and un- are provided as follows. binding rates, respectively; U and B denote the unbound and Reaction-Diffusion Models with Time-varying Input: bound states of the receptor, respectively. When there are NR One of the first attempts to model the ligand-receptor binding receptors, assuming that all of them are exposed to the same reactions from an MC theoretical perspective is provided concentration of ligands, reaction rate equation (RRE) for the in [165], where the authors develop a noise model for the number of bound receptors can be written as [165] fluctuations in the number of bound receptors of a receiver dn (t) exposed to time-varying ligand concentrations as MC signals. B = k c (t)(N − n (t)) − k n (t). (15) dt + L R B − B The model is based on the assumption of a spherical receiver, in which ligand receptors and information-carrying ligands are As is clear, while the binding reaction is second-order de- homogeneously distributed. For an analytically tractable anal- pending on the concentrations of both ligands and available ysis, the concentration of incoming ligands is assumed to be receptors, unbinding reaction is first-order and only depends constant between two sampling times, i.e., during a sampling on the number of bound receptors. interval, and the ligand-receptor binding reaction is assumed to Most of the time, the bandwidth of MC signals can be be at equilibrium at the beginning of each sampling interval. In assumed to be low enough to drive the binding reaction light of these assumptions, the authors obtain the time-varying to near equilibrium and allow applying quasi steady-state variance and mean of the number of bound receptors, which assumption for the overall system. In this case, time-varying are valid only for the corresponding sampling interval. A more concentration c (t) can be treated constant, i.e., c (t) = c , L L L general approach without the equilibrium assumption to obtain and dn (t)/dt = 0, which results in the following expression B the mean number of bound receptors with time-varying input for the mean number of bound receptors signals, i.e., ligand concentration, is contributed by [284] and cL E[nB] = NR, (16) [167], through solving the system of differential equations cL + KD governing the overall diffusion-reaction MC system. The au- where KD = k−/k+ is the dissociation constant, which is thors of the both studies consider a spherical receiver with a measure of affinity between the specific type of ligand and ligand receptors on its surface and a point transmitter, which receptor. Even at equilibrium, the receptors randomly fluctuate can be anywhere on a virtual sphere centered at the same between the bound and unbound state. The number of bound point as the receiver but larger than that, to obtain a spherical receptors nB, at equilibrium, is a Binomial random variable symmetry to simplify the overall problem. As a result, the with success probability pB = cL/(cL +KD), and its variance transmitter location cannot be exactly specified in the problem. can be given accordingly by In [167], the authors consider that the spherical receiver is capable of binding ligands at any point on its surface, which is Var[n ] = p (1 − p )N . B B B R (17) exactly equal to the assumption of infinite number of receptors. More insight can be gained by examining the continuous On the other hand, [284] considers finite number of receptors history of binding and unbinding events over receptors. The uniformly distributed on the receiver surface, and addresses likelihood of observing a series of n binding-unbinding events this challenge through boundary homogenization. However, at equilibrium can be given by boundary homogenization for finite number of receptors does not take into account the negative feedback of the bound n M B U 1 T PM k+c Y X + − −k−τ B receptors on the second-order binding reaction (see (15)), and p({τ , τ } ) = e U i=1 i i k c k e i j , n Z i i i thus, the developed analytical model is not able to capture the j=1 i=1 (18) indirect effects of finite number of receptors, e.g., receptor saturation. This is clear from their analysis, such that the Pn U where Z is the normalization factor, TU = j=1 τj is the discrepancy between the analytical model and the particle- U B th total unbound time, τj and τj are the j unbound and based simulation results is getting larger with increasing ligand + − bound time intervals, respectively, ci, ki and ki are the concentration. concentration, binding rate, and unbinding rate of ith type of Frequency Domain Model: Another modeling approach ligand, respectively, M is the number of ligand types present is provided in [285], where the authors, assuming that the in the channel [282], [283]. Note that the likelihood is equally probability of a receptor to be in the bound state is very low, valid for the cases of single receptor and multiple receptors, take the number of available, i.e., unbound, receptors equal to 26 the total number of receptors at all time points. The completely interplay between convection, diffusion and reaction is taken first order characteristics of the resulting RRE enables them into account by defining a transport-modified reaction rate, to carry out a frequency domain analysis, through which tailored for the hemicylindrical surface of the SiNW bioFET they show the ligand-receptor binding reaction manifests low- receiver. However, the authors assume steady-state conditions pass filter characteristics. However, this approximate model is for the reaction, to be able to derive a closed-form expression relevant only when the probability of receptor-ligand binding for the noise statistics. The molecular-to-electrical transduction is very low. properties of the bioFET are reflected to the output current of Discrete Model based on Reaction-Diffusion Master the receiver through modeling the capacitive effects arising Equation (RDME): To capture the stochasticity of the from the liquid-semiconductor interface and the 1/f noise reaction-diffusion MC, another approach is introduced in resulting from the defects of the SiNW transducer channel. In [286], where the authors develop a voxel-based model based [288], the authors considered a two-dimensional convection- on RDME, with the diffusion and reactions at the receiver reaction-diffusion system, which does not lend itself to closed- modeled as Markov processes. The three-dimensional MC form analytical expressions for the received signal. Authors system is discretized and divided into equal-size cubic voxels, develop a heuristic model using a two-compartmental model- in each of which molecules are assumed to be uniformly ing approach, which divides the channel into compartments, in distributed, and allowed to move only to the neighboring each of which either transport or reaction occurs, and derive voxels. In the voxel accommodating the Tx, the molecules an analytical expression for the time-course of the number are generated according to a modulation scheme, and the Rx of bound receptors over a planar receiver surface placed at voxel hosts the receptor molecules, where the ligands diffusing the bottom of the channel. The model well captures the into the Rx voxel can react based on law of mass action. nonlinearities, such as Taylor-Aris dispersion in the channel, The jump of a ligand from one voxel to another is governed depletion region above the receiver surface, and saturation by a diffusion rate parameter, which is a function of the effects resulting from finite number of receptors, as validated voxel size and the ligand diffusion coefficient. The number of through finite element simulations of the system in COMSOL. ligands and bound receptors are stored in a system state vector, However, the model assumes the channel and the receiver is which is progressed with a given state transition rate vector empty at the beginning of the transmission; therefore, does not storing the reaction, diffusion, and molecule generation rates. allow an ISI analysis. In the continuum limit, the model is able to provide closed- b) Detection Methods: The literature on detection meth- form analytical expressions for the mean and variance of the ods for MC with reactive receivers is relatively scarce, and number of bound receptors for small-scale systems. However, the reason can be attributed partly to the lack of analytical for larger systems, with a high number of voxels, the efficiency models that can capture the nonlinear ligand-receptor binding of the model is highly questionable. reaction kinetics and resulting noise and ISI. Nevertheless, Steady-state Model: In addition to the above approaches the existing methods can be divided into three categories considering time-varying signals, some researchers prefer us- depending on the type of assumptions made and considered ing the assumption of steady-state ligand-receptor binding receiver architectures. reaction with stationary input signals at the time of sampling, Detection based on Instantaneous Receptor States: The based on fact that the bandwidth of incoming MC signals is first detection approach is based on sampling the instantaneous typically low because the diffusion channel shows low-pass number of bound receptors at a prespecified time, as shown in filter characteristics and the reaction rates are generally higher Fig. 11, and comparing it to a threshold. In [167], the authors than the diffusion rate of molecules. This assumption enables study a threshold-based detection for OOK modulated ligand the separation of the overall system into two; a deterministic concentrations, using the difference between the number of microscale diffusion channel and the stochastic ligand-receptor bound molecules at the start and end of a bit interval. In binding reaction at the interface between the receiver and the [285], converting the ligand-receptor binding reaction to a channel. Accordingly, at the sampling time, the ligand concen- completely first-order reaction with the assumption that all tration around the receptors assumes different constant values of the receptors are always available for binding, the authors corresponding to different symbols. The only fluctuations are manage to transform the problem into the frequency domain. resulting from the binding reaction, where the random number For the modulation, they consider MoSK with different recep- of bound receptors follows Binomial distribution, whose mean tors corresponding to different ligands; therefore, the problem and variance are given in (16) and (17), respectively. The basically reduces to a detection problem of the concentration- steady-state assumption is applied in [166], [287], where the encoded signals for each ligand-receptor pair. To reduce the authors derive reaction-diffusion channel capacity for different amount of noise, they propose to apply a whitening filter to settings. the sensed signal in the form of number of bound receptors, Convection-Diffusion-Reaction System Model: Microflu- and then utilize the same detection technique they proposed idic MC systems with reactive receivers are studied by a few in [268]. An energy-based detection scheme is proposed in researchers. In [178], authors develop a one-dimensional an- [289], where the test statistics is the total number of binding alytical model, assuming that the propagation occurs through events that occur within a symbol duration. They propose convection and diffusion, and a reactive receiver, which is a variable threshold-detection scheme with varying memory assumed to be a SiNW bioFET receiver with ligand receptors length. The article also takes into account the ISI; however, on its surface, is placed at the bottom of the channel. The the model assumes that all the receptors are always available 27 for binding, and completely neglects the unbinding of ligands Receptor state Receptor from the receptors, making the reaction irreversible. In [283], bound time intervals the authors study the saturation problem in MC-Rxs with Bound ligand receptors. As part of their analysis, they investigate the Receptor 1 Unbound performance of an adaptive threshold ML detection using the Time instantaneous number of bound receptors. The effect of the Bound receiver memory that stores the previously decoded bits, is also Receptor 2 Unbound investigated, and their analysis reveal that the performance of Time MC detection based on number of bound receptors severely . decreases when the receptors get saturated, which occurs when . they are exposed to a high concentration of ligands as a result Bound Receptor NR of strong ISI. This is because near saturation, it becomes Unbound harder for the receiver to discriminate between two levels of Time Receptor Instantaneous number of bound receptors corresponding to different bits. unbound time receptor states Detection based on Continuous History of Receptor intervals States: The second detection approach is based on exploit- Fig. 11: Different types of sampling of receptor states in ing the continuous history of binding and unbinding events reactive MC receivers. occurring at receptors, or the independent samples of time intervals that the receptors stay bound and unbound, as demon- strated in Fig. 11. As we see in the likelihood function in (18), the unbound time intervals are informative of the total performance of the one-shot ML detection schemes based on ligand concentration, whereas the bound time intervals are number of bound receptors and unbound time intervals in the informative of the type of bound molecules. This is exploited presence of interference from a similar ligand available in the in [283], where the authors tackle the saturation problem in environment, number of which in the reception space follows reactive receivers. For a receiver with ligand receptors and Poisson distribution, with a known mean. They proposed a with a memory storing a finite number of previously decoded new ML detection scheme based on estimating the ratio of bits, they propose to exploit the amount of time the individ- messenger ligands to the inferring ligands using the bound ual receptors stay unbound. Taking the ligand concentration time intervals sampled from each receptor. It is shown that stationary around the sampling time, and assuming steady- the proposed method substantially outperforms the others, state conditions for the ligand-receptor binding reaction, they especially when the concentration of interferer molecules is developed an ML detection scheme for OOK concentration- very high. Note that the above detection techniques are built encoded signals, which outperforms the detection based on on MC models that neglect the receiver geometry by treating number of bound receptors in the saturation case. A simple it as a transparent receiver, inside which ligands and receptors intracellular reaction network to perform the transduction of are homogeneously distributed, and they require the receiver unbound time intervals into the concentration of certain type to store an internal model, i.e., CIR, of the reaction-diffusion of molecules inside the cell is also designed. In a similar channel. manner, in [290], [291], using a voxel-based MC system model introduced in [286], and by neglecting the ISI, the authors Detection for Biosensor-based MC Receivers: The third developed an optimal MAP demodulator scheme based on set of detection methods deal with the biosensor-based re- the continuous history of receptor binding events. Different ceivers, where the binding events are transduced into elec- from [283], the authors assume time-varying input signal. The trical signals. In bioFET-based receivers the concentration of resulting demodulator is an analog filter, which requires the bound charge-carrying ligands are converted into electrical biochemical implementation of mathematical operations, such signals that are contaminated with additional noise. It is not as logarithm, multiplication, and integration. The demodulator possible to observe individual receptors states; therefore, the also needs to count the number of binding events. In [290], detection based on continuous history of binding events is not they provide an extension of the demodulator for the ISI applicable for these receivers. Accounting for the 1/f noise case by incorporating a decision feedback, and show the and binding fluctuations at steady-state conditions, in [178], performance improvement with increasing receiver memory. authors develop an optimal ML detection scheme for CSK in One of the challenges of reactive receivers is their selectivity the absence of ISI. Approximating the binding and 1/f noise towards the messenger molecules, which is not perfect in with a Gaussian distribution, they reduce the overall problem practice. It is highly probable, especially in physiologically to a fixed-threshold detection problem and provide closed- relevant environments, that there are similar ligands in the form analytical expressions for the optimal thresholds and channel, which can also bind the same receptors, even though corresponding symbol error rates. The performance evaluation their unbinding rate is higher than that of the correct, i.e., reveals that the 1/f noise, which is resulting from the defects messenger, ligands. This causes a molecular interference, of the semiconductor FET channel, surpasses binding noise which impedes the detection performance of the receiver resulting from the fluctuations of the receptor states, espe- especially when the concentration of interferer molecules is cially at low frequencies, and severely degrades the detection not known to the receiver. In [292], the authors evaluate the performance. 28

IV. CHALLENGESAND FUTURE RESEARCH DIRECTIONS MC-Tx and MC-Rx, the harvestable energy can be quite After providing a comprehensive survey of existing studies limited so that complex algorithms may not be feasible on different aspects of MC, in this section, we discuss the most in a realistic MC scenario. important challenges toward realization of micro/nanoscale • Data Rate: MC is mostly promising in applications MC setups and evaluation of their ICT-based performance. In without high data rate requirements as MC suffers from this direction, we highlight both the required theoretical studies slow propagation channels. This problem can be tackled and the experimental investigations to validate the theoretical by encoding large amount of data in DNA/RNA strands, models. Physical architecture of MC-Tx/Rx is one of the least which is named as NSK as discussed in Section II-C. studied topics in the MC literature; however, it significantly This way, the amount of transmitted information can affects the accuracy of theoretical studies. Thus, we provide be increased up to 100s of MB per IM such that MC an in-depth discussion on the required investigations of MC can achieve data rates on the order of Mbps. Although Tx/Rx architecture. The second shortcoming of MC literature sequencing of DNA/RNA strands can be performed via is use of non-realistic assumptions in the existing MC-Tx/Rx standalone devices utilizing nanopores, there is yet any and channel models, which has led to imprecise performance practical low-cost system to write DNA sequences with estimations not applicable to real scenarios. Hence, we explain a micro/nanoscale device. the requirements of realistic modeling, such as taking into • Molecule Leakage: In the case of nanomaterial-based account the impact of stochastic molecule generation process, MC-Txs, there are two significant design challenges: Tx/Rx geometry, and stochastic noise dynamics. We also molecule leakage and molecule reservoir/generation. describe open research problems in developed modulation, Molecule leakage, while transmitting low logic or no coding and detection techniques for MC and emphasize im- information, is unavoidable for practical MC-Tx designs. portant factors that must be considered in future investigations. The leaking molecules increase ISI in the channel, and also contribute to an additional problem in molecule reservoir/generation by lowering the molecule budget of A. Challenges for Physical Design and Implementation of MC-Tx. In order to tackle this problem, we have proposed MC-Tx some solutions based on molecule wax as an IM container Theoretical modeling of MC-Tx in the literature generally and hybrid MC-Tx designs, where genetically engineered relies on unrealistic simplifying assumptions due to the lack bacteria can be utilized to replenish IM sources. How- of experimental studies on micro/nanoscale implementation of ever, these solutions have not been implemented and the MC-Tx. Therefore, MC-Tx is often assumed as an ideal point feasibility of such systems as MC-Tx still stands as an source capable of perfectly transmitting molecular messages, important open research issue. which completely neglects important factors in the IM release • Biological Complexity: In the case of biological MC-Tx process such as the stochasticity in the molecule generation architectures, the main difficulty in design stems from process, the effect of Tx geometry and channel feedback. As complexity of involved biological elements. Understand- modeling micro/nanoscale MC-Tx without any experimental ing of molecular basis underlying cellular mechanisms, data seems not possible, requirement for empirical transmitter i.e., cellular biology, and development of techniques to models is the most pressing challenge for MC as end-to- manipulate them, i.e., synthetic biology, are essential to end channel models are required to consider the effects of unlock the reliable use of engineered biological entities both MC-Tx and MC-Rx. Based on the empirical channel for MC. Moreover, as in many applications data propa- models, the communication theoretical investigation of MC, gated in nanonetworks eventually needs to be connected i.e., analysis of capacity, modulation, coding and detection, to electronic devices, any biological network architecture needs to be revisited as communication parameters show must be interfaced with an electronic architecture. Thus, strong dependence on the channel model. Other challenges the development of MC architectures within this space regarding the practical implementation of a micro/nanoscale requires an extremely high interdisciplinary engagement MC-Tx are listed as follows between the fields of ICT, biology and synthetic biology, • Energy: Micro/nanoscale MC-Tx and MC-Rx are re- as well as materials science and nanofabrication. Another quired to work as standalone devices with their own en- issue, which again is caused by the inherent complexity ergy sources. Since battery-powered devices have limited of biological organisms, is the fact that, genetically engi- lifetime, EH techniques are promising to develop energy neered cells are not the best survivors, and complications neutral devices such that all operations of the device can in their engineered metabolisms cause accelerated death be powered via harvested energy from various sources of the cell. such as solar, mechanical and chemical. To design such systems, one needs to first calculate the harvestable en- ergy budget of an MC-Tx and MC-Rx, i.e., the amount of B. Challenges for Physical Design and Implementation of energy harvestable from the surrounding based on appli- MC-Rx cation scenarios and medium. Then, both transmitter and Despite existing theoretical studies on the performance of receiver operations, e.g., modulation, coding, detection, MC [11], [159]–[163], [165]–[167] and few macroscale exper- are required to be developed accordingly not to exceed imental setups [83], [179], [180], [257], [258], the literature the available energy budget. Considering miniaturized lacks comprehensive investigation of the physical design of 29 micro/nano-scale MC-Rx structures. This leads to critical open actor networks inside human body are two promising challenges needed to be tackled for realization of the MC examples of these applications. To this aim, implementa- promising applications. tion of micro/nanoscale bio-cyber interfaces are required. The bio-cyber interface needs to decode the molecular • Ligand-Receptor Selection: As mentioned in Section messages, process it and send the decoded information III-C, both genetic circuit based and artificial architec- to a macroscale network node through a wireless link. tures use ligand-receptor reactions to sense the con- In bioFET-based MC-Rxs, thanks to the molecular-to- centration of target molecule by the MC antenna. This electrical transducer unit, the electrical signal generated calls for careful selection of appropriate ligand-receptor by the receiver can be sent to the cyber-networks through pairs for MC paradigm. In this regards, the binding EM wireless communications. However, connection of and dissociation, i.e., unbinding, rates of these reactions biological MC-Rxs to the cyber-networks remains as an are among the important parameters that must be taken open issue. into consideration. The binding rate controls sensitivity, Lastly, it is worthwhile to mention that the physical archi- selectivity and the response time of the device. While tecture of the device is mainly dictated by the application high dissociation rates reduces the re-usability of the requirements. As an example, the biological MC-Rxs are only device, very low values are also not desired as they promising and feasible for in vivo applications. On the other decrease the sensitivity of the receiver. The existing hand, utilizing artificial structures for in vivo applications interferer molecules in the application environment and necessitates comprehensive investigation of the receiver bio- their affinities with the receptors are the next important compatibility. Moreover, the physiological conditions imply design parameter that must be studied to minimize the solutions with high ionic concentrations, abundance of in- background noise. terferers and contaminants, and existence of disruptive flows • Realistic ICT-based modeling of artificial structures: and fluctuating temperature, which may degrade the receivers For artificial biosensor-based MC-Rxs, the literature is performance in several aspects. First, high ionic concentration lacking analytical models that can capture transient dy- creates strong screening effect reducing the Debye length, thus, namics, as the available models developed from sensor impedes the sensitivity of the receiver [234]. Moreover, con- application perspective are mostly based on equilibrium taminants and disruptive flows may alter the binding kinetics, assumption. However, in MC applications, since the con- impede the stability of the receptors, even separate them from centration signals are time-varying, equilibrium models the dielectric layer. These call for comprehensive investigation may not be realistic. Therefore, devising MC detection of these factors, their impact on the performance of the device methods for artificial receivers requires developing more and methods to control their effects. To control screening prob- complex models that can also capture the stochastic lem, using highly charged ligands and very small size receptors noise dynamics without steady-state assumption for dif- like aptamers are theoretically efficient. However, frequency ferent types of biosensors based on nanomaterials, e.g., domain technique promises for much more realistic solutions. graphene bioFET. The models should also include the In [294], the authors reveal that frequency domain detection effects of operating voltages for bioFETs, receiver geom- outperforms the conventional time domain technique in terms etry and gate configuration, channel ionic concentration of sensitivity in highly ionic solutions. An alternative solution determining the strength of Debye screening. The models to overcome the Debye screening limitations in detection is should be validated through wet-lab experiments, and proposed in [295], where it is shown that applying a high- microfluidic platforms stand as a promising option for frequency alternating current between the source and the drain implementing testbeds for MC systems with artificial electrodes, instead of a DC current, weakens the double layer MC-Rxs. capacitance generated by the solution ions. They show that the • Biological Circuits Complexity: In the case of imple- effective charges of the ligands become inversely dependent on menting MC-Rxs with genetic circuits, the information the Debye length instead of the exponential dependence. As a transmission is through molecules and biochemical reac- result, the screening effect on the bound ligands is significantly tions, which results in nonlinear input-output behaviors reduced, and the FET-based biosensing becomes feasible even with system-evolution-dependent stochastic effects. This for physiological conditions. Similar approaches are taken by makes the analytically studying the performance metrics others to overcome the Debye screening with radio-frequency difficult, if not impossible. Moreover, the existing noise operation of graphene bioFETs, which are summarized in in genetic circuits must be comprehensively studied and [296]. methods to mitigate this noise must be derived as it significantly reduces the achievable mutual information of the MC [175]. C. Challenges for Developing MC Modulation Techniques • Bio-Cyber Interface: While MC expands the function- There are various modulation schemes that are proposed for ality of nanomachines by connecting them to each other MC by utilizing concentration, molecule type/order/ratio, and and making nanonetworks [293], connection of these release timing. However, these studies mostly utilize simplified nanonetworks to the cyber-networks, i.e., making IoBNT, channel models based on MC-Tx, which is an ideal point further extend the applications of these nanomachines source, and MC-Rx, which is capable of perfectly detecting [1]. Continuous health monitoring and bacterial sensor- multiple molecules selectively. Therefore, the performance of 30 the proposed schemes under realistic conditions are still un- transmitter and the receiver. In fact, synchronization is es- known, and this can be tackled by implementing the proposed sential for the proper operation of the proposed solutions. schemes in an experimental MC setup. In addition, some of There are many studies proposing different synchroniza- the modulation schemes require synchronization, which is hard tion methods, e.g., using quorum sensing to globally syn- to achieve considering the error-prone diffusive channel and chronize the actions of nanomachines in a nanonetwork low complexity MC devices at micro/nanoscale. Furthermore, [299], blind synchronization and clock synchronization energy efficiency is another important challenge for MC-Tx based on the ML estimation of the channel delay [300], designs as being powered via limited energy harvested from [301], peak observation-time-based and threshold-trigger- the medium. Therefore, energy-efficient and low complexity based symbol synchronization schemes employing a dif- modulation schemes for MC without strict synchronization ferent type of molecule for the purpose of synchronization requirements still stand as a significant research problem. [302]. However, these methods are either too complex for the limited capabilities for the nanomachines, or they D. Challenges for Developing MC Channel Coding Tech- rely on stable CSI, which is not the case for time- niques varying MC channels. Moreover, the effect of these non- Even though there is some literature on MC channel coding ideal synchronization techniques on the performance of techniques, the field is still very new and open for research. the proposed detection methods has not been revealed. Some of the most pressing directions are highlighted below. Another potential solution to the synchronization problem could be to develop asynchronous detection techniques • Adaptive Coding: The propagation time through MC that obviate the need for synchronization. As reviewed channel and detection probabilities at the receiver are in Section III-D1, there are a few promising solutions affected by various environmental parameters such as for asynchronous MC detection, e.g., peak-detection and temperature, diffusion speed, channel contents, reaction threshold-based detection methods [276]. However, they rates and distance between nodes, which calls for adaptive are mostly built on simplifying assumptions for receiver channel coding techniques for error compensation [297]. and channel geometry and properties, which may result In this respect, none of the coding schemes investigated in unexpected performance in real applications. even considers to probe the channel for its characteristics • Physical Properties of the Receiver: Although there in order to adapt itself. is little physical correspondence for passive and absorb- • Irregular Signaling: All of the coding schemes dis- ing receivers, many of the detection schemes are built cussed have been evaluated under the regular time-slotted on these assumptions, as they enable a mathematically transmission assumption, which in any realistic MC sce- tractable analysis. As reviewed in Section III-D2, al- nario will never be the case. For instance, fluctuations in though they consider the effect of receptor reactions, the the inter-symbol duration caused by irregular transmis- initial studies on reactive receivers also follow similar sion would have a significant effect on suffered ISI, the assumptions on the device architecture and geometry to primary source of BER in MC, and therefore needs to be simplify the analyses. However, for practical systems, considered by channel codes. the physical properties of the realistic receiver archi- • Simplicity of Models: More channel codes for MC need tectures and their impact on the molecular propagation to be invented. The only works that do so are [145]–[147], in the MC channel should be taken into consideration and they have very simple transmission and channel to the most possible extent. Finite element simulations models, i.e., in all works the transmitter releases only a on microfluidic MC channel with bioFET receivers and single molecule per transmission period, and the channel macroscale MC experiments with alcohol sensors clearly is one-dimensional. reveal the effect of the coupling between the MC receiver and the channel [288], [298]. The coupling is highly E. Challenges for Developing MC Detection Techniques nonlinear, and in most of the cases, it is not analytically As reviewed in Section III-D, the MC detection problem tractable; therefore, beyond the available analytical tools, has been widely addressed from several aspects for different researchers may need to focus on stochastic simulations channel and receiver configurations. However, the proposed and experiments to validate the performance of the pro- solutions are still far from being feasible for envisioned MC posed detection techniques in realistic scenarios. devices. The main reason behind this discrepancy between • Physical Properties of the Channel: Most of the MC the theoretical solutions and the real practice, which is also detection studies assume free diffusion, or diffusion plus revealed by the preliminary experimental airborne MC studies uniform flow, for molecular propagation in an unbounded performed with macroscale off-the-shelf components [83], 3D environment. However, in practice MC channel will [298], is that there is no micro/nanoscale MC testbed that can be bounded with varying boundary conditions, and can be used as a validation framework to optimize the devised include nonuniform and disruptive flows, obstacles, tem- methods. In parallel to this general problem regarding MC perature fluctuations, and charged molecules affecting technologies, major challenges for MC detection can be further the diffusion coefficient, and particles leading to channel detailed as follows: crowding. Some of these aspects have recently started to • Synchronization: The majority of the proposed detection be addressed through channel modeling studies, e.g., sub- techniques assume a perfect synchronization between the diffusive MC channel due to molecular crowding [303], 31

diffusion-based MC in multilayered biological media present in the channel can bind the same receptors as the [304]. However, these simplified models are not able messenger ligands, causing molecular interference. Even to provide enough insight into the detection problem in if there are different types of receptor molecules for each realistic channels. The highly time-varying properties of ligand type in an MC system with MoSK, the cross-talk the MC channel have also been addressed by researchers between receptors is unavoidable because of the non- through channel estimation techniques [305], [306] and ideal coupling between receptor and ligands. Therefore, noncoherent detection techniques [272]–[274], which are there is a need for selective detection methods exploiting reviewed in Section III-D1. These studies are built on the properties of ligand-receptor binding reaction. The simplifying assumptions on channel and receiver archi- amount of time a receptor stays bound is informative of tecture. the ligand unbinding rate, which is directly linked with • Reactive Receivers: Although reactive receiver concept the affinity between a particular type of ligand-receptor provides a more realistic approach to the detection prob- pair. The bound time intervals are exploited in [292] to lem, the research in this direction is still at its infancy, detect the MC messages based on the ML estimation of and the developed received signal models are not complex the ratio of correct ligands to the interferers. However, enough to reflect many intricacies. For example, most of this technique requires the receiver to know the type and the previous studies assume independent receptors being probability distribution of the interferer molecules, and it exposed to the same ligand concentration; however, this is developed only for one type of interferer. To implement is not always the case, as the binding of one receptor selective detection methods in engineered bacteria-based can affect the binding of neighboring ones [307], and MC devices, the kinetic proofreading and adaptive sorting receptors can form cooperative clusters to control sen- techniques implemented by reaction networks of the T sitivity by exploiting spatial heterogeneity [308], [309]. cells in the immune system can be exploited [282], [313], The interplay between diffusion and reaction is also often [314]. Additionally, MC spectrum sensing methods can neglected in these studies by assuming the timescales be developed based on the information inferred from of both processes are separated sufficiently. However, the receptor bound time intervals to apply cognitive in most practical cases, reaction and diffusion rates are radio techniques in crowded MC nanonetworks where close to each other, and reactions are correlated with the many nanomachines communicate using same type of transport process, which depends on the channel proper- molecules. ties [230], [310]. These problems are crucial to analyze the spatio-temporal correlations among the receptors and V. CONCLUSION the coupling between the diffusion channel and reac- MC has attracted significant research attention over the tive receiver. Moreover, except for a few recent studies last decade. Although ICT aspects of MC, i.e., information [154], [311], [312], the design of intracellular reaction theoretical models of MC channels, modulation and detection networks to implement the proposed detectors are usually techniques for MC and system-theoretical modeling of MC neglected. The additional noise and delay stemming from applications, are well-studied, these research efforts mostly these reaction networks should be taken into account rely on unrealistic assumptions isolating the MC channel from while evaluating the performance of the overall detection. the physical processes regarding transmitting and receiving Furthermore, a proper analysis of the trade-off between operations. The reason being that although there are some energy, detection accuracy and detection speed is required proposals for MC-Tx/Rx based on synthetic biology and to develop an optimization framework for the detector nanomaterials, there is no implementation of any artificial design in reactive receivers. micro/nanoscale MC system to date. As a result, feasibility • Receiver Saturation: Another challenge associated with and performance of ICT techniques proposed for MC could reactive receivers is the saturation of receptors in the case not be validated. of strong ISI or external interference, which can severely In this paper, we provide a comprehensive survey on the limit the receiver dynamic range, and hamper the ability recent proposals for the physical design of MC-Tx/Rx and of the receiver to discriminate different signal levels. The the state-of-the-art in the theoretical MC research covering saturation problem has been recently addressed in [283] modulation, coding and detection techniques. We first investi- through steady-state assumption for the ligand-receptor gate the fundamental requirements for the physical design of binding reactions; however, in general it is neglected micro/nanoscale MC-Tx/Rx in terms of energy and molecule assuming an infinite number of receptors on the receiver. consumption, and operating conditions. In light of these re- The problem can also be alleviated by adaptive threshold quirements, the state-of-the-art design approaches as well as detection techniques and adaptive transmission schemes novel MC-Tx/Rx architectures, i.e., artificial Tx/Rx designs as in detection with passive/absorbing receivers. enabled by the nanomaterials, e.g., graphene, and biological • Receiver Selectivity: Receiver selectivity is also a ma- Tx/Rx designs enabled by synthetic biology, are covered. jor issue for MC detection, and it has just started to In addition, we highlight the opportunities and challenges be addressed from MC perspective. The physiologically regarding the implementation of Tx/Rx and corresponding ICT relevant environments usually include many similar types techniques which are to be built on these devices. of molecules, and the receptor-ligand coupling is not The guidelines on the physical design of MC-Tx/Rxs that typically ideal, such that many different types of ligands are provided in this paper will help the researchers to design 32

experimental MC setups and develop realistic channel models [21] K. Srinivas, A. W. Eckford, and R. S. Adve, “Molecular communication considering the transceiving processes. In this way, the long- in fluid media: The additive inverse gaussian noise channel,” IEEE transactions on information theory, vol. 58, no. 7, p. 4678, 2012. standing discrepancy between theory and practice in MC [22] L. Organick, S. D. Ang, Y.-J. Chen, R. Lopez, S. Yekhanin, can be overcome towards unleashing the huge societal and K. Makarychev, M. Z. Racz, G. Kamath, P. Gopalan, B. Nguyen et al., economic impact of MC by paving the way for ICT-based “Random access in large-scale dna data storage,” Nature biotechnology, vol. 36, no. 3, p. 242, 2018. early diagnosis and treatment of diseases, such as IoNT-based [23] C. Berrou, A. Glavieux, and P. Thitimajshima, “Near shannon limit continuous health monitoring, smart drug delivery, artificial error-correcting coding and decoding: Turbo-codes. 1,” in Communi- organs, and lab-on-a-chips. cations, 1993. ICC’93 Geneva. Technical Program, Conference Record, IEEE International Conference on, vol. 2. IEEE, 1993, pp. 1064–1070. [24] R. W. Hamming, “Error detecting and error correcting codes,” Bell REFERENCES System technical journal, vol. 29, no. 2, pp. 147–160, 1950. [25] D. Daksh, D. Rawtani, and Y. K. Agrawal, “Recent developments [1] I. F. Akyildiz, M. Pierobon, S. Balasubramaniam, and Y. Koucheryavy, in bio-nanoelectronics devices: A review,” Journal of Bionanoscience, “The internet of bio-nano things,” IEEE Communications Magazine, vol. 10, no. 2, pp. 81–93, 2016. vol. 53, no. 3, pp. 32–40, 2015. [26] Q. H. Abbasi, K. Yang, N. Chopra, J. M. Jornet, N. A. Abuali, [2] I. F. Akyildiz and J. M. Jornet, “The internet of nano-things,” IEEE K. A. Qaraqe, and A. Alomainy, “Nano-communication for biomedical Wireless Communications, vol. 17, no. 6, 2010. applications: A review on the state-of-the-art from physical layers to [3] O. B. Akan, H. Ramezani, T. Khan, N. A. Abbasi, and M. Kuscu, novel networking concepts,” IEEE Access, vol. 4, pp. 3920–3935, 2016. “Fundamentals of molecular information and communication science,” [27] R. B. Schoch, J. Han, and P. Renaud, “Transport phenomena in Proceedings of the IEEE, vol. 105, no. 2, pp. 306–318, 2017. nanofluidics,” Reviews of modern physics, vol. 80, no. 3, p. 839, 2008. [4] L. Felicetti, M. Femminella, G. Reali, and P. Lio,` “Applications of [28] I. F. Akyildiz, F. Brunetti, and C. Blazquez,´ “Nanonetworks: A new molecular communications to medicine: A survey,” Nano Communica- communication paradigm,” Computer Networks, vol. 52, no. 12, pp. tion Networks, vol. 7, pp. 27–45, 2016. 2260–2279, 2008. [5] B. Atakan, O. B. Akan, and S. Balasubramaniam, “Body area nanonet- [29] J.-T. Huang and C.-H. Lee, “On capacity bounds of two-way diffusion works with molecular communications in nanomedicine,” IEEE Com- channel with molecule harvesting,” in Communications (ICC), 2017 munications Magazine, vol. 50, no. 1, 2012. IEEE International Conference on. IEEE, 2017, pp. 1–6. [6] N. Farsad, H. B. Yilmaz, A. Eckford, C.-B. Chae, and W. Guo, “A [30] H. G. Bafghi, A. Gohari, M. Mirmohseni, and M. Nasiri-Kenari, comprehensive survey of recent advancements in molecular communi- “Diffusion based molecular communication with limited molecule cation,” IEEE Communications Surveys & Tutorials, vol. 18, no. 3, pp. production rate,” arXiv preprint arXiv:1802.08965, 2018. 1887–1919, 2016. [31] B. D. Unluturk, A. O. Bicen, and I. F. Akyildiz, “Genetically engi- [7] S. Qiu, W. Guo, S. Wang, N. Farsad, and A. Eckford, “A molecular neered bacteria-based biotransceivers for molecular communication,” communication link for monitoring in confined environments,” in Com- IEEE Transactions on Communications, vol. 63, no. 4, pp. 1271–1281, munications Workshops (ICC), 2014 IEEE International Conference on. 2015. IEEE, 2014, pp. 718–723. [32] B. A. Bilgin and O. B. Akan, “A Fast Algorithm for Analysis of [8] M. Pierobon and I. F. Akyildiz, “Capacity of a diffusion-based molecu- Molecular Communication in Artificial Synapse,” IEEE Transactions lar communication system with channel memory and molecular noise,” on Nanobioscience, vol. 16, no. 6, pp. 408–417, 2017. IEEE Transactions on Information Theory, vol. 59, no. 2, pp. 942–954, [33] C. Pinero-Lambea,˜ D. Ruano-Gallego, and L. A.´ Fernandez,´ “Engi- 2013. neered bacteria as therapeutic agents,” Current opinion in biotechnol- [9] B. Atakan and O. B. Akan, “Deterministic capacity of information flow ogy, vol. 35, pp. 94–102, 2015. in molecular nanonetworks,” Nano Communication Networks, vol. 1, [34] S. Haefner, P. Frank, E. Langer, D. Gruner, U. Schmidt, M. Elstner, no. 1, pp. 31–42, 2010. G. Gerlach, and A. Richter, “Chemically controlled micro-pores and [10] M. S. Kuran, H. B. Yilmaz, T. Tugcu, and I. F. Akyildiz, “Modulation nano-filters for separation tasks in 2d and 3d microfluidic systems,” techniques for communication via diffusion in nanonetworks,” in Com- RSC Advances, vol. 7, no. 78, pp. 49 279–49 289, 2017. munications (ICC), 2011 IEEE International Conference on. IEEE, [35] C. Di Natale, G. Celetti, P. L. Scognamiglio, C. Cosenza, E. Battista, 2011, pp. 1–5. F. Causa, and P. A. Netti, “Molecularly endowed hydrogel with an [11] D. Kilinc and O. B. Akan, “Receiver design for molecular communi- in silico-assisted screened peptide for highly sensitive small molecule cation,” IEEE Journal on Selected Areas in Communications, vol. 31, harvesting,” Chemical Communications, vol. 54, no. 72, pp. 10 088– no. 12, pp. 705–714, 2013. 10 091, 2018. [12] Y. Chahibi, “Molecular communication for drug delivery systems: A [36] E. Katz, Implantable bioelectronics. John Wiley & Sons, 2014. survey,” Nano Communication Networks, vol. 11, pp. 90–102, 2017. [37] S. Sharifi, S. Behzadi, S. Laurent, M. L. Forrest, P. Stroeve, and [13] R. Mosayebi, A. Ahmadzadeh, W. Wicke, V. Jamali, R. Schober, and M. Mahmoudi, “Toxicity of nanomaterials,” Chemical Society Reviews, M. Nasiri-Kenari, “Early cancer detection in blood vessels using mobile vol. 41, no. 6, pp. 2323–2343, 2012. nanosensors,” arXiv preprint arXiv:1805.08777, 2018. [38] F. Asghari, M. Samiei, K. Adibkia, A. Akbarzadeh, and S. Davaran, [14] S. Murdan, “Electro-responsive drug delivery from hydrogels,” Journal “Biodegradable and biocompatible polymers for tissue engineering ap- of controlled release, vol. 92, no. 1-2, pp. 1–17, 2003. plication: a review,” Artificial cells, nanomedicine, and biotechnology, [15] A. Scott, K. Weir, C. Easton, W. Huynh, W. J. Moody, and A. Folch, “A vol. 45, no. 2, pp. 185–192, 2017. microfluidic microelectrode array for simultaneous electrophysiology, [39] J. M. Anderson, “Mechanisms of inflammation and infection with chemical stimulation, and imaging of brain slices,” Lab on a Chip, implanted devices,” Cardiovascular Pathology, vol. 2, no. 3, pp. 33–41, vol. 13, no. 4, pp. 527–535, 2013. 1993. [16] A. Jonsson, S. Inal, I. Uguz, A. J. Williamson, L. Kergoat, J. Rivnay, [40] T. Someya, Z. Bao, and G. G. Malliaras, “The rise of plastic bioelec- D. Khodagholy, M. Berggren, C. Bernard, G. G. Malliaras et al., tronics,” Nature, vol. 540, no. 7633, p. 379, 2016. “Bioelectronic neural pixel: Chemical stimulation and electrical sensing [41] C. K. Nguyen, N. Q. Tran, T. P. Nguyen, and D. H. Nguyen, “Bio- at the same site,” Proceedings of the National Academy of Sciences, compatible nanomaterials based on dendrimers, hydrogels and hydrogel vol. 113, no. 34, pp. 9440–9445, 2016. nanocomposites for use in biomedicine,” Advances in Natural Sciences: [17] M. Kuscu and O. B. Akan, “On the physical design of molecular Nanoscience and Nanotechnology, vol. 8, no. 1, p. 015001, 2017. communication receiver based on nanoscale biosensors,” IEEE Sensors [42] T. M. Allen and P. R. Cullis, “Drug delivery systems: entering the Journal, vol. 16, no. 8, pp. 2228–2243, 2016. mainstream,” Science, vol. 303, no. 5665, pp. 1818–1822, 2004. [18] R. Daniel, J. R. Rubens, R. Sarpeshkar, and T. K. Lu, “Synthetic analog [43] D. T. Simon, S. Kurup, K. C. Larsson, R. Hori, K. Tybrandt, M. Goiny, computation in living cells,” Nature, vol. 497, no. 7451, p. 619, 2013. E. W. Jager, M. Berggren, B. Canlon, and A. Richter-Dahlfors, “Or- [19] O. Purcell and T. K. Lu, “Synthetic analog and digital circuits for ganic electronics for precise delivery of neurotransmitters to modulate cellular computation and memory,” Current opinion in biotechnology, mammalian sensory function,” Nature materials, vol. 8, no. 9, p. 742, vol. 29, pp. 146–155, 2014. 2009. [20] M. S¸. Kuran, H. B. Yilmaz, T. Tugcu, and I. F. Akyildiz, “Interference [44] T. S. Moon, C. Lou, A. Tamsir, B. C. Stanton, and C. A. Voigt, “Genetic effects on modulation techniques in diffusion based nanonetworks,” programs constructed from layered logic gates in single cells,” Nature, Nano Communication Networks, vol. 3, no. 1, pp. 65–73, 2012. vol. 491, no. 7423, p. 249, 2012. 33

[45] P. Siuti, J. Yazbek, and T. K. Lu, “Synthetic circuits integrating logic [69] Y. Wang, D. Zheng, Q. Tan, M. X. Wang, and L.-Q. Gu, “Nanopore- and memory in living cells,” Nature biotechnology, vol. 31, no. 5, p. based detection of circulating micrornas in lung cancer patients,” 448, 2013. Nature nanotechnology, vol. 6, no. 10, p. 668, 2011. [46] W. H. Bossert and E. O. Wilson, “The analysis of olfactory commu- [70] S. L. Shipman, J. Nivala, J. D. Macklis, and G. M. Church, “Crispr–cas nication among animals,” Journal of theoretical biology, vol. 5, no. 3, encoding of a digital movie into the genomes of a population of living pp. 443–469, 1963. bacteria,” Nature, vol. 547, no. 7663, p. 345, 2017. [47] L. C. Cobo and I. F. Akyildiz, “Bacteria-based communication in [71] D. E. Clapham, “Calcium signaling,” Cell, vol. 131, no. 6, pp. 1047– nanonetworks,” Nano Communication Networks, vol. 1, no. 4, pp. 244– 1058, 2007. 256, 2010. [72] J. Isaksson, P. Kjall,¨ D. Nilsson, N. Robinson, M. Berggren, and [48] M. Pierobon and I. F. Akyildiz, “A physical end-to-end model for A. Richter-Dahlfors, “Electronic control of ca 2+ signalling in neuronal molecular communication in nanonetworks,” IEEE Journal on Selected cells using an organic electronic ion pump,” Nature materials, vol. 6, Areas in Communications, vol. 28, no. 4, 2010. no. 9, p. 673, 2007. [49] F. Walsh, S. Balasubramaniam, D. Botvich, T. Suda, T. Nakano, [73] T. Nakano, T. Suda, M. Moore, R. Egashira, A. Enomoto, and S. F. Bush, and M. O.´ Foghlu,´ “Hybrid dna and enzyme based K. Arima, “Molecular communication for nanomachines using in- computing for address encoding, link switching and error correction tercellular calcium signaling,” in Nanotechnology, 2005. 5th IEEE in molecular communication,” in International Conference on Nano- Conference on. IEEE, 2005, pp. 478–481. Networks. Springer, 2008, pp. 28–38. [74] I. Uguz, C. M. Proctor, V. F. Curto, A.-M. Pappa, M. J. Donahue, [50] S. Hiyama, Y. Moritani, R. Gojo, S. Takeuchi, and K. Sutoh, M. Ferro, R. M. Owens, D. Khodagholy, S. Inal, and G. G. Malliaras, Advanced Mate- “Biomolecular-motor-based autonomous delivery of lipid vesicles as “A microfluidic ion pump for in vivo drug delivery,” rials nano-or microscale reactors on a chip,” Lab on a Chip, vol. 10, no. 20, , vol. 29, no. 27, p. 1701217, 2017. pp. 2741–2748, 2010. [75] A. Jonsson, T. A. Sjostr¨ om,¨ K. Tybrandt, M. Berggren, and D. T. Simon, “Chemical delivery array with millisecond neurotransmitter [51] A. Enomoto, M. J. Moore, T. Suda, and K. Oiwa, “Design of self- release,” Science advances, vol. 2, no. 11, p. e1601340, 2016. organizing microtubule networks for molecular communication,” Nano [76] M. C. Lee, E. A. Miller, J. Goldberg, L. Orci, and R. Schekman, “Bi- Communication Networks, vol. 2, no. 1, pp. 16–24, 2011. directional protein transport between the er and golgi,” Annu. Rev. Cell [52] M. Gregori, I. Llatser, A. Cabellos-Aparicio, and E. Alarcon,´ “Physical Dev. Biol., vol. 20, pp. 87–123, 2004. channel characterization for medium-range nanonetworks using flagel- [77] K. Denzer, M. J. Kleijmeer, H. Heijnen, W. Stoorvogel, and H. J. lated bacteria,” Computer Networks, vol. 55, no. 3, pp. 779–791, 2011. Geuze, “Exosome: from internal vesicle of the multivesicular body to [53] O. B. Akan, O. Cetinkaya, C. Koca, and M. Ozger, “Internet of hybrid intercellular signaling device,” Journal of cell science, vol. 113, no. 19, energy harvesting things,” IEEE Internet of Things Journal, vol. 5, pp. 3365–3374, 2000. no. 2, pp. 736–746, 2018. [78] S. R. Schwarze, A. Ho, A. Vocero-Akbani, and S. F. Dowdy, “In vivo [54] C. Dagdeviren, Z. Li, and Z. L. Wang, “Energy harvesting from the protein transduction: delivery of a biologically active protein into the animal/human body for self-powered electronics,” Annual review of mouse,” Science, vol. 285, no. 5433, pp. 1569–1572, 1999. biomedical engineering, vol. 19, pp. 85–108, 2017. [79] F. Walsh, S. Balasubramaniam, D. Botvich, and W. Donnelly, “Syn- [55] V. Leonov, “Thermoelectric energy harvesting of human body heat for thetic protocols for nano sensor transmitting platforms using enzyme wearable sensors,” IEEE Sensors Journal, vol. 13, no. 6, pp. 1–8, 2013. and dna based computing,” Nano Communication Networks, vol. 1, [56] M. Amin Karami and D. J. Inman, “Powering pacemakers from no. 1, pp. 50–62, 2010. heartbeat vibrations using linear and nonlinear energy harvesters,” [80] M. Chourasia and S. Jain, “Pharmaceutical approaches to colon targeted Applied Physics Letters, vol. 100, no. 4, p. 042901, 2012. drug delivery systems,” J Pharm Pharm Sci, vol. 6, no. 1, pp. 33–66, [57] W. Jia, G. Valdes-Ram´ ´ırez, A. J. Bandodkar, J. R. Windmiller, and 2003. J. Wang, “Epidermal biofuel cells: energy harvesting from human per- [81] K. Cho, X. Wang, S. Nie, D. M. Shin et al., “Therapeutic nanoparticles spiration,” Angewandte Chemie International Edition, vol. 52, no. 28, for drug delivery in cancer,” Clinical cancer research, vol. 14, no. 5, pp. 7233–7236, 2013. pp. 1310–1316, 2008. [58] S. Hiyama and Y. Moritani, “Molecular communication: Harnessing [82] N.-R. Kim and C.-B. Chae, “Novel modulation techniques using biochemical materials to engineer biomimetic communication sys- isomers as messenger molecules for nano communication networks tems,” Nano Communication Networks, vol. 1, no. 1, pp. 20–30, 2010. via diffusion,” IEEE Journal on Selected Areas in Communications, [59] C. J. Donnelly, M. Fainzilber, and J. L. Twiss, “Subcellular commu- vol. 31, no. 12, pp. 847–856, 2013. nication through rna transport and localized protein synthesis,” Traffic, [83] N. Farsad, W. Guo, and A. W. Eckford, “Tabletop molecular commu- vol. 11, no. 12, pp. 1498–1505, 2010. nication: Text messages through chemical signals,” PloS one, vol. 8, [60] M. Mittelbrunn and F. Sanchez-Madrid,´ “Intercellular communication: no. 12, p. e82935, 2013. diverse structures for exchange of genetic information,” Nature reviews [84] M. Prakash and N. Gershenfeld, “Microfluidic bubble logic,” Science, Molecular cell biology, vol. 13, no. 5, p. 328, 2012. vol. 315, no. 5813, pp. 832–835, 2007. [61] N. A. Bell and U. F. Keyser, “Digitally encoded dna nanostructures for [85] M. J. Fuerstman, P. Garstecki, and G. M. Whitesides, “Coding/decoding multiplexed, single-molecule protein sensing with nanopores,” Nature and reversibility of droplet trains in microfluidic networks,” Science, nanotechnology, vol. 11, no. 7, pp. 645–651, 2016. vol. 315, no. 5813, pp. 828–832, 2007. [86] S. L. Peterson, A. McDonald, P. L. Gourley, and D. Y. Sasaki, “Poly [62] K. Chen, M. Juhasz, F. Gularek, E. Weinhold, Y. Tian, U. F. Keyser, and (dimethylsiloxane) thin films as biocompatible coatings for microfluidic N. A. Bell, “Ionic current-based mapping of short sequence motifs in devices: cell culture and flow studies with glial cells,” Journal of single dna molecules using solid-state nanopores,” Nano letters, vol. 17, Biomedical Materials Research Part A: An Official Journal of The no. 9, pp. 5199–5205, 2017. Society for Biomaterials, The Japanese Society for Biomaterials, and [63] E. Slonkina and A. B. Kolomeisky, “Polymer translocation through a The Australian Society for Biomaterials and the Korean Society for The Journal of chemical physics long nanopore,” , vol. 118, no. 15, pp. Biomaterials, vol. 72, no. 1, pp. 10–18, 2005. 7112–7118, 2003. [87] N. Farsad, A. W. Eckford, S. Hiyama, and Y. Moritani, “On-chip [64] N. A. Bell, M. Muthukumar, and U. F. Keyser, “Translocation fre- molecular communication: Analysis and design,” IEEE Transactions quency of double-stranded dna through a solid-state nanopore,” Phys- on NanoBioscience, vol. 11, no. 3, pp. 304–314, 2012. ical Review E, vol. 93, no. 2, p. 022401, 2016. [88] E. De Leo, L. Donvito, L. Galluccio, A. Lombardo, G. Morabito, [65] N. A. Bell and U. F. Keyser, “Direct measurements reveal non- and L. M. Zanoli, “Communications and switching in microfluidic markovian fluctuations of dna threading through a solid-state systems: Pure hydrodynamic control for networking labs-on-a-chip,” nanopore,” arXiv preprint arXiv:1607.04612, 2016. IEEE transactions on communications, vol. 61, no. 11, pp. 4663–4677, [66] T. Z. Butler, J. H. Gundlach, and M. A. Troll, “Determination of 2013. rna orientation during translocation through a biological nanopore,” [89] R. Iezzi, P. Finlayson, Y. Xu, and R. Katragadda, “Microfluidic Biophysical Journal, vol. 90, no. 1, pp. 190–199, 2006. neurotransmiter-based neural interfaces for retinal prosthesis,” in Engi- [67] C. Dekker, “Solid-state nanopores,” Nature nanotechnology, vol. 2, neering in Medicine and Biology Society, 2009. EMBC 2009. Annual no. 4, p. 209, 2007. International Conference of the IEEE. IEEE, 2009, pp. 4563–4565. [68] S. Hernandez-Ainsa´ and U. F. Keyser, “Dna origami nanopores: devel- [90] P. D. Jones and M. Stelzle, “Can nanofluidic chemical release enable opments, challenges and perspectives,” Nanoscale, vol. 6, no. 23, pp. fast, high resolution neurotransmitter-based neurostimulation?” Fron- 14 121–14 132, 2014. tiers in neuroscience, vol. 10, p. 138, 2016. 34

[91] M. I. Walker, K. Ubych, V. Saraswat, E. A. Chalklen, P. Braeuninger- [112] C. E. Dunbar, K. A. High, J. K. Joung, D. B. Kohn, K. Ozawa, and Weimer, S. Caneva, R. S. Weatherup, S. Hofmann, and U. F. Keyser, M. Sadelain, “Gene therapy comes of age,” Science, vol. 359, no. 6372, “Extrinsic cation selectivity of 2d membranes,” ACS nano, vol. 11, p. eaan4672, 2018. no. 2, pp. 1340–1346, 2017. [113] J. Yu, J. Moffitt, C. L. Hetherington, C. Bustamante, and G. Oster, [92] A. Fabbro, D. Scaini, V. Leon, E. Vazquez, G. Cellot, G. Privitera, “Mechanochemistry of a viral dna packaging motor,” Journal of L. Lombardi, F. Torrisi, F. Tomarchio, F. Bonaccorso et al., “Graphene- molecular biology, vol. 400, no. 2, pp. 186–203, 2010. based interfaces do not alter target nerve cells,” ACS nano, vol. 10, [114] F. Walsh and S. Balasubramaniam, “Reliability of multi-path virus no. 1, pp. 615–623, 2016. nanonetworks,” in Communications Workshops (ICC), 2013 IEEE [93] Y. Brudno, E. A. Silva, C. J. Kearney, S. A. Lewin, A. Miller, K. D. International Conference on. IEEE, 2013, pp. 824–828. Martinick, M. Aizenberg, and D. J. Mooney, “Refilling drug delivery [115] ——, “Reliability and delay analysis of multihop virus-based nanonet- depots through the blood,” Proceedings of the National Academy of works,” IEEE Transactions on Nanotechnology, vol. 12, no. 5, pp. 674– Sciences, vol. 111, no. 35, pp. 12 722–12 727, 2014. 684, 2013. [94] Y. Brudno, M. J. Pezone, T. K. Snyder, O. Uzun, C. T. Moody, [116] S. Mathivanan, H. Ji, and R. J. Simpson, “Exosomes: extracellular M. Aizenberg, and D. J. Mooney, “Replenishable drug depot to combat organelles important in intercellular communication,” Journal of pro- post-resection cancer recurrence,” Biomaterials, 2018. teomics, vol. 73, no. 10, pp. 1907–1920, 2010. [95] W. Whyte, E. T. Roche, C. E. Varela, K. Mendez, S. Islam, H. ONeill, [117] S. Abadal and I. F. Akyildiz, “Automata modeling of quorum sensing F. Weafer, R. N. Shirazi, J. C. Weaver, N. V. Vasilyev et al., “Sustained for nanocommunication networks,” Nano Communication Networks, release of targeted cardiac therapy with a replenishable implanted vol. 2, no. 1, pp. 74–83, 2011. epicardial reservoir,” Nature Biomedical Engineering, vol. 2, no. 6, [118] S. Abadal, I. Llatser, E. Alarcon,´ and A. Cabellos-Aparicio, “Quorum p. 416, 2018. sensing-enabled amplification for molecular nanonetworks,” in Com- [96] D. Hanahan, “Studies on transformation of escherichia coli with munications (ICC), 2012 IEEE International Conference on. IEEE, plasmids,” Journal of molecular biology, vol. 166, no. 4, pp. 557–580, 2012, pp. 6162–6166. 1983. [119] T. Nakano, M. J. Moore, F. Wei, A. V. Vasilakos, and J. Shuai, “Molec- [97] G. J. Hannon, “Rna interference,” nature, vol. 418, no. 6894, p. 244, ular communication and networking: Opportunities and challenges,” 2002. IEEE transactions on nanobioscience, vol. 11, no. 2, pp. 135–148, [98] J. L. Mellies and E. R. Lawrence-Pine, “Interkingdom signaling 2012. between pathogenic bacteria and caenorhabditis elegans,” Trends in [120] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “On the character- microbiology, vol. 18, no. 10, pp. 448–454, 2010. ization of binary concentration-encoded molecular communication in [99] S. Xiang, J. Fruehauf, and C. J. Li, “Short hairpin rna–expressing nanonetworks,” Nano Communication Networks, vol. 1, no. 4, pp. 289– bacteria elicit rna interference in mammals,” Nature biotechnology, 300, 2010. vol. 24, no. 6, p. 697, 2006. [121] N. Garralda, I. Llatser, A. Cabellos-Aparicio, E. Alarcon,´ and M. Pier- [100] D. T. Riglar and P. A. Silver, “Engineering bacteria for diagnostic and obon, “Diffusion-based physical channel identification in molecular therapeutic applications,” Nature Reviews Microbiology, vol. 16, no. 4, nanonetworks,” Nano Communication Networks, vol. 2, no. 4, pp. 196– p. 214, 2018. 204, 2011. [101] R. McKay, M. Ghodasra, J. Schardt, D. Quan, A. E. Pottash, W. Shang, [122] M. H. Kabir, S. R. Islam, and K. S. Kwak, “D-mosk modulation S. M. Jay, G. F. Payne, M. W. Chang, J. C. March et al., “A platform in molecular communications,” IEEE transactions on nanobioscience, of genetically engineered bacteria as vehicles for localized delivery of vol. 14, no. 6, pp. 680–683, 2015. therapeutics: Toward applications for crohn’s disease,” Bioengineering [123] Y. Murin, N. Farsad, M. Chowdhury, and A. Goldsmith, “Exploiting & Translational Medicine, 2018. diversity in molecular timing channels via order statistics,” arXiv [102] A. M. Sugranes˜ and I. F. Akyildiz, “Capacity and delay of bacteria- preprint arXiv:1801.05567, 2018. based communication in nanonetworks,” Ph.D. dissertation, Universitat [124] ——, “Communication over diffusion-based molecular timing chan- Politecnica` de Catalunya. Escola Tecnica` Superior d’Enginyeria de nels,” in Global Communications Conference (GLOBECOM), 2016 Telecomunicacio´ de Barcelona. Departament d’Arquitectura de Com- IEEE. IEEE, 2016, pp. 1–6. putadors, 2012 (Enginyeria de Telecomunicacio),´ 2012. [125] B. Atakan, S. Galmes, and O. B. Akan, “Nanoscale communica- [103] S. Balasubramaniam et al., “Opportunistic routing through conjuga- tion with molecular arrays in nanonetworks,” IEEE Transactions on tion in bacteria communication nanonetwork,” Nano Communication NanoBioscience, vol. 11, no. 2, pp. 149–160, 2012. Networks, vol. 3, no. 1, pp. 36–45, 2012. [126] M. Brini, T. Cal`ı, D. Ottolini, and E. Carafoli, “Intracellular calcium [104] ——, “Multi-hop conjugation based bacteria nanonetworks,” IEEE homeostasis and signaling,” in Metallomics and the Cell. Springer, Transactions on nanobioscience, vol. 12, no. 1, pp. 47–59, 2013. 2013, pp. 119–168. [105] S. Balasubramaniam, N. Lyamin, D. Kleyko, M. Skurnik, A. Vinel, [127] M. L. Cotrina, J. H.-C. Lin, A. Alves-Rodrigues, S. Liu, J. Li, H. Azmi- and Y. Koucheryavy, “Exploiting bacterial properties for multi-hop Ghadimi, J. Kang, C. C. Naus, and M. Nedergaard, “Connexins regulate nanonetworks,” IEEE Communications Magazine, vol. 52, no. 7, pp. calcium signaling by controlling atp release,” Proceedings of the 184–191, 2014. National Academy of Sciences, vol. 95, no. 26, pp. 15 735–15 740, [106] V. Petrov, D. Moltchanov, and Y. Koucheryavy, “Analytical model for 1998. link reliability in bacteria nanonetworks,” in Proceedings of ACM The [128] T. Nakano, T. Suda, T. Koujin, T. Haraguchi, and Y. Hiraoka, “Molec- First Annual International Conference on Nanoscale Computing and ular communication through gap junction channels: System design, Communication. ACM, 2014, p. 17. experiments and modeling,” in Bio-Inspired Models of Network, In- [107] V. Petrov, D. Moltchanov, S. Balasubramaniam, and Y. Koucheryavy, formation and Computing Systems, 2007. Bionetics 2007. 2nd. IEEE, “Incorporating bacterial properties for plasmid delivery in nano sensor 2007, pp. 139–146. networks,” IEEE Transactions on Nanotechnology, vol. 14, no. 4, pp. [129] T. Nakano and J.-Q. Liu, “Design and analysis of molecular relay 751–760, 2015. channels: An information theoretic approach,” IEEE Transactions on [108] V. Petrov, S. Balasubramaniam, R. Lale, D. Moltchanov, Y. Kouch- NanoBioscience, vol. 9, no. 3, pp. 213–221, 2010. eryavy et al., “Forward and reverse coding for chromosome transfer in [130] M. T. Barros, S. Balasubramaniam, B. Jennings, and Y. Koucheryavy, bacterial nanonetworks,” Nano Communication Networks, vol. 5, no. “Transmission protocols for calcium-signaling-based molecular com- 1-2, pp. 15–24, 2014. munications in deformable cellular tissue,” IEEE Transactions on [109] V. Petrov, B. Deng, D. Moltchanov, S. Balasubramaniam, and Nanotechnology, vol. 13, no. 4, pp. 779–788, 2014. Y. Koucheryavy, “Performance comparison of message encoding tech- [131] M. T. Barros, S. Balasubramaniam, and B. Jennings, “Using infor- niques for bacterial nanonetworks,” in Wireless Communications and mation metrics and molecular communication to detect cellular tissue Networking Conference (WCNC), 2016 IEEE. IEEE, 2016, pp. 1–7. deformation,” IEEE transactions on nanobioscience, vol. 13, no. 3, pp. [110] T. Furubayashi, T. Nakano, A. Eckford, Y. Okaie, and T. Yomo, “Packet 278–288, 2014. fragmentation and reassembly in molecular communication,” IEEE [132] ——, “Comparative end-to-end analysis of ca 2+-signaling-based transactions on nanobioscience, vol. 15, no. 3, pp. 284–288, 2016. molecular communication in biological tissues,” IEEE Transactions on [111] P. Palese, H. Zheng, O. G. Engelhardt, S. Pleschka, and A. Garc´ıa- Communications, vol. 63, no. 12, pp. 5128–5142, 2015. Sastre, “Negative-strand rna viruses: genetic engineering and appli- [133] M. T. Barros, “Ca2+-signaling-based molecular communication sys- cations,” Proceedings of the National Academy of Sciences, vol. 93, tems: Design and future research directions,” Nano Communication no. 21, pp. 11 354–11 358, 1996. Networks, vol. 11, pp. 103–113, 2017. 35

[134] M. Moore, A. Enomoto, T. Nakano, R. Egashira, T. Suda, A. Kayasuga, [155] ——, “A biological circuit design for modulated parity-check encoding H. Kojima, H. Sakakibara, and K. Oiwa, “A design of a molecular in molecular communication,” in Communications (ICC), 2017 IEEE communication system for nanomachines using molecular motors,” in International Conference on. IEEE, 2017, pp. 1–7. Pervasive Computing and Communications Workshops, 2006. PerCom [156] D. A. Armbruster and T. Pry, “Limit of blank, limit of detection and Workshops 2006. Fourth Annual IEEE International Conference on. limit of quantitation,” The Clinical Biochemist Reviews, vol. 29, no. IEEE, 2006, pp. 6–pp. Suppl 1, p. S49, 2008. [135] A. Enomoto, M. Moore, T. Nakano, R. Egashira, T. Suda, A. Kayasuga, [157] P. R. Nair and M. A. Alam, “Design considerations of silicon nanowire H. Kojima, H. Sakakibara, and K. Oiwa, “A molecular communication biosensors,” IEEE Transactions on Electron Devices, vol. 54, no. 12, system using a network of cytoskeletal filaments,” in Proceedings of pp. 3400–3408, 2007. the 2006 NSTI Nanotechnology Conference, 2006. [158] K. Kalantar-Zadeh, Sensors: an introductory course. Springer Science [136] A. W. Eckford, N. Farsad, S. Hiyama, and Y. Moritani, “Microchannel & Business Media, 2013. molecular communication with nanoscale carriers: Brownian motion [159] L.-S. Meng, P.-C. Yeh, K.-C. Chen, and I. F. Akyildiz, “On receiver versus active transport,” in Nanotechnology (IEEE-NANO), 2010 10th design for diffusion-based molecular communication,” IEEE Transac- IEEE Conference on. IEEE, 2010, pp. 854–858. tions on Signal Processing, vol. 62, no. 22, pp. 6032–6044, 2014. [137] N. Farsad, A. W. Eckford, and S. Hiyama, “Design and optimizing [160] A. Noel, K. C. Cheung, and R. Schober, “Optimal receiver design for of on-chip kinesin substrates for molecular communication,” IEEE diffusive molecular communication with flow and additive noise,” IEEE Transactions on Nanotechnology, vol. 14, no. 4, pp. 699–708, 2015. transactions on nanobioscience, vol. 13, no. 3, pp. 350–362, 2014. [138] S. Sankaran, J. Becker, C. Wittmann, and A. del Campo, “Optoregu- [161] R. Mosayebi, H. Arjmandi, A. Gohari, M. Nasiri-Kenari, and U. Mitra, lated drug release from an engineered living hydrogel,” 2018. “Receivers for diffusion-based molecular communication: Exploiting [139] S. R. Lindemann, H. C. Bernstein, H.-S. Song, J. K. Fredrickson, M. W. memory and sampling rate,” IEEE Journal on Selected Areas in Fields, W. Shou, D. R. Johnson, and A. S. Beliaev, “Engineering mi- Communications, vol. 32, no. 12, pp. 2368–2380, 2014. crobial consortia for controllable outputs,” The ISME journal, vol. 10, [162] H. B. Yilmaz, A. C. Heren, T. Tugcu, and C.-B. Chae, “3-d chan- no. 9, p. 2077, 2016. nel characteristics for molecular communications with an absorbing [140] Y. Murin, M. Chowdhury, N. Farsad, and A. Goldsmith, “Diversity gain receiver,” IEEE Communications Letters, vol. 18, no. 6, pp. 929–932, of one-shot communication over molecular timing channels,” arXiv 2014. preprint arXiv:1709.07976, 2017. [163] M. Damrath, S. Korte, and P. A. Hoeher, “Equivalent discrete-time [141] M. Movahednasab, M. Soleimanifar, A. Gohari, M. Nasiri-Kenari, and channel modeling for molecular communication with emphasize on U. Mitra, “Adaptive transmission rate with a fixed threshold decoder an absorbing receiver,” IEEE transactions on nanobioscience, vol. 16, for diffusion-based molecular communication,” IEEE Transactions on no. 1, pp. 60–68, 2017. Communications, vol. 64, no. 1, pp. 236–248, 2016. [164] E. Dinc and O. B. Akan, “Theoretical limits on multiuser molecular [142] M. Blawat, K. Gaedke, I. Hutter,¨ X.-M. Chen, B. Turczyk, S. Inverso, communication in internet of nano-bio things,” IEEE transactions on B. W. Pruitt, and G. M. Church, “Forward error correction for dna data nanobioscience, vol. 16, no. 4, pp. 266–270, 2017. Procedia Computer Science storage,” , vol. 80, pp. 1011–1022, 2016. [165] M. Pierobon and I. F. Akyildiz, “Noise analysis in ligand-binding [143] J. Clarke, H.-C. Wu, L. Jayasinghe, A. Patel, S. Reid, and H. Bay- reception for molecular communication in nanonetworks,” IEEE Trans- ley, “Continuous base identification for single-molecule nanopore dna actions on Signal Processing, vol. 59, no. 9, pp. 4168–4182, 2011. sequencing,” Nature nanotechnology, vol. 4, no. 4, p. 265, 2009. [166] G. Aminian, M. Farahnak-Ghazani, M. Mirmohseni, M. Nasiri-Kenari, [144] I. M. Derrington, T. Z. Butler, M. D. Collins, E. Manrao, M. Pavlenok, and F. Fekri, “On the capacity of point-to-point and multiple-access M. Niederweis, and J. H. Gundlach, “Nanopore dna sequencing with molecular communications with ligand-receptors.” T-MBMC, vol. 1, mspa,” Proceedings of the National Academy of Sciences, vol. 107, no. 4, pp. 331–346, 2015. no. 37, pp. 16 060–16 065, 2010. [167] Y. Deng, A. Noel, M. Elkashlan, A. Nallanathan, and K. C. Cheung, [145] P.-J. Shih, C.-h. Lee, and P.-C. Yeh, “Channel codes for mitigating in- “Modeling and simulation of molecular communication systems with tersymbol interference in diffusion-based molecular communications,” a reversible adsorption receiver,” IEEE Transactions on Molecular, in Global Communications Conference (GLOBECOM), 2012 IEEE. Biological and Multi-Scale Communications, vol. 1, no. 4, pp. 347– IEEE, 2012, pp. 4228–4232. 362, 2015. [146] P.-J. Shih, C.-H. Lee, P.-C. Yeh, and K.-C. Chen, “Channel codes for reliability enhancement in molecular communication,” IEEE Journal [168] M. Pierobon, “A systems-theoretic model of a biological circuit for Nano Communication on Selected Areas in Communications, vol. 31, no. 12, pp. 857–867, molecular communication in nanonetworks,” Networks 2013. , vol. 5, no. 1-2, pp. 25–34, 2014. [147] P.-Y. Ko, Y.-C. Lee, P.-C. Yeh, C.-h. Lee, and K.-C. Chen, “A new [169] C. J. Myers, Engineering genetic circuits. Chapman and Hall/CRC, paradigm for channel coding in diffusion-based molecular communica- 2016. tions: Molecular coding distance function,” in Global Communications [170] D. Baker, G. Church, J. Collins, D. Endy, J. Jacobson, J. Keasling, Conference (GLOBECOM), 2012 IEEE. IEEE, 2012, pp. 3748–3753. P. Modrich, C. Smolke, and R. Weiss, “Engineering life: building a [148] M. S. Leeson and M. D. Higgins, “Forward error correction for fab for biology,” Scientific American, vol. 294, no. 6, pp. 44–51, 2006. molecular communications,” Nano Communication Networks, vol. 3, [171] R. Sarpeshkar, “Analog synthetic biology,” Phil. Trans. R. Soc. A, vol. no. 3, pp. 161–167, 2012. 372, no. 2012, p. 20130110, 2014. [149] C. Bai, M. S. Leeson, and M. D. Higgins, “Minimum energy channel [172] N. Arkus, “A mathematical model of cellular apoptosis and senescence codes for molecular communications,” Electronics Letters, vol. 50, through the dynamics of telomere loss,” Journal of theoretical biology, no. 23, pp. 1669–1671, 2014. vol. 235, no. 1, pp. 13–32, 2005. [150] Y. Lu, M. D. Higgins, and M. S. Leeson, “Self-orthogonal convolu- [173] P. K. Maini, “Using mathematical models to help understand biological tional codes (soccs) for diffusion-based molecular communication sys- pattern formation,” Comptes rendus biologies, vol. 327, no. 3, pp. 225– tems,” in Communications (ICC), 2015 IEEE International Conference 234, 2004. on. IEEE, 2015, pp. 1049–1053. [174] E. Klipp and W. Liebermeister, “Mathematical modeling of intracellular [151] Y. Lu, X. Wang, M. D. Higgins, A. Noel, N. Neophytou, and M. S. Lee- signaling pathways,” BMC neuroscience, vol. 7, no. 1, p. S10, 2006. son, “Energy requirements of error correction codes in diffusion-based [175] C. Harper, M. Pierobon, and M. Mazarini, “Estimating information molecular communication systems,” Nano Communication Networks, exchange performance of engineered cell-to-cell molecular communi- vol. 11, pp. 24–35, 2017. cations: a computational approach,” in IEEE INFOCOM 2018-IEEE [152] Y. Lu, M. D. Higgins, and M. S. Leeson, “Comparison of channel Conference on Computer Communications. IEEE, 2018, pp. 729– coding schemes for molecular communications systems,” IEEE Trans- 737. actions on Communications, vol. 63, no. 11, pp. 3991–4001, 2015. [176] T. Ramalho, A. Meyer, A. Muckl,¨ K. Kapsner, U. Gerland, and F. C. [153] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “Performance analysis Simmel, “Single cell analysis of a bacterial sender-receiver system,” of convolutional coding techniques in diffusion-based concentration- PloS one, vol. 11, no. 1, p. e0145829, 2016. encoded pam molecular communication systems,” BioNanoScience, [177] F. W. Scheller, U. Wollenberger, A. Warsinke, and F. Lisdat, “Research vol. 3, no. 3, pp. 270–284, 2013. and development in biosensors,” Current Opinion in Biotechnology, [154] A. Marcone, M. Pierobon, and M. Magarini, “Parity-check coding vol. 12, no. 1, pp. 35–40, 2001. based on genetic circuits for engineered molecular communication [178] M. Kuscu and O. B. Akan, “Modeling and analysis of sinw fet-based between biological cells,” IEEE Transactions on Communications, molecular communication receiver,” IEEE Transactions on Communi- 2018. cations, vol. 64, no. 9, pp. 3708–3721, 2016. 36

[179] N. Farsad, D. Pan, and A. Goldsmith, “A novel experimental platform ethanol detection,” ACS applied materials & interfaces, vol. 9, no. 16, for in-vessel multi-chemical molecular communications,” in GLOBE- pp. 14 051–14 057, 2017. COM 2017-2017 IEEE Global Communications Conference. IEEE, [201] S. Eissa and M. Zourob, “In vitro selection of dna aptamers targeting 2017, pp. 1–6. β-lactoglobulin and their integration in graphene-based biosensor for [180] H. Unterweger, J. Kirchner, W. Wicke, A. Ahmadzadeh, D. Ahmed, the detection of milk allergen,” Biosensors and Bioelectronics, vol. 91, V. Jamali, C. Alexiou, G. Fischer, and R. Schober, “Experimental pp. 169–174, 2017. molecular communication testbed based on magnetic nanoparticles in [202] E. Luzi, M. Minunni, S. Tombelli, and M. Mascini, “New trends duct flow,” arXiv preprint arXiv:1803.06990, 2018. in affinity sensing: aptamers for ligand binding,” TrAC Trends in [181] S. M. Borisov and O. S. Wolfbeis, “Optical Biosensors,” Chemical Analytical Chemistry, vol. 22, no. 11, pp. 810–818, 2003. Reviews, vol. 108, no. 2, pp. 423–461, feb 2008. [Online]. [203] A. Ganguly, C.-P. Chen, Y.-T. Lai, C.-C. Kuo, C.-W. Hsu, K.-H. Chen, Available: http://www.ncbi.nlm.nih.gov/pubmed/18229952http://pubs. and L.-C. Chen, “Functionalized gan nanowire-based electrode for acs.org/doi/abs/10.1021/cr068105t direct label-free voltammetric detection of dna hybridization,” Journal [182] J. Arlett, E. Myers, and M. Roukes, “Comparative advantages of of Materials Chemistry, vol. 19, no. 7, pp. 928–933, 2009. mechanical biosensors,” Nature nanotechnology, vol. 6, no. 4, p. 203, [204] W. Yang, K. R. Ratinac, S. P. Ringer, P. Thordarson, J. J. Gooding, 2011. and F. Braet, “Carbon nanomaterials in biosensors: should you use [183] J. Tamayo, P. M. Kosaka, J. J. Ruz, A.´ San Paulo, and M. Calleja, nanotubes or graphene?” Angewandte Chemie International Edition, “Biosensors based on nanomechanical systems,” Chemical Society vol. 49, no. 12, pp. 2114–2138, 2010. Reviews, vol. 42, no. 3, pp. 1287–1311, 2013. [205] D. Sarkar, W. Liu, X. Xie, A. C. Anselmo, S. Mitragotri, and [184] J. Wang, “Electrochemical glucose biosensors,” Chemical reviews, vol. K. Banerjee, “Mos2 field-effect transistor for next-generation label-free 108, no. 2, pp. 814–825, 2008. biosensors,” ACS nano, vol. 8, no. 4, pp. 3992–4003, 2014. [185] K. R. Rogers, “Principles of affinity-based biosensors,” Molecular [206] L. Torsi, M. Magliulo, K. Manoli, and G. Palazzo, “Organic field-effect biotechnology, vol. 14, no. 2, pp. 109–129, 2000. transistor sensors: a tutorial review,” Chemical Society Reviews, vol. 42, [186] Y. Huang, X. Dong, Y. Shi, C. M. Li, L.-J. Li, and P. Chen, “Nano- no. 22, pp. 8612–8628, 2013. electronic biosensors based on cvd grown graphene,” Nanoscale, vol. 2, [207] M. Curreli, R. Zhang, F. N. Ishikawa, H.-K. Chang, R. J. Cote, C. Zhou, no. 8, pp. 1485–1488, 2010. and M. E. Thompson, “Real-time, label-free detection of biological [187] A. Poghossian and M. J. Schoning,¨ “Label-free sensing of biomolecules entities using nanowire-based fets,” IEEE Trans. Nanotechnol, vol. 7, with field-effect devices for clinical applications,” Electroanalysis, no. 6, pp. 651–667, 2008. vol. 26, no. 6, pp. 1197–1213, 2014. [208] S. U. Senveli and O. Tigli, “Biosensors in the small scale: methods [188] M. F. Bear, B. W. Connors, and M. A. Paradiso, Neuroscience:exploring and technology trends,” IET nanobiotechnology, vol. 7, no. 1, pp. 7– the brain. Lippincott Williams & Wilkins, 2007, vol. 2. 21, 2013. [189] J. Shan, J. Li, X. Chu, M. Xu, F. Jin, X. Wang, L. Ma, X. Fang, [209] D. P. Tran, M. A. Winter, B. Wolfrum, R. Stockmann, C.-T. Yang, Z. Wei, and X. Wang, “High sensitivity glucose detection at extremely M. Pourhassan-Moghaddam, A. Offenhausser, and B. Thierry, “Toward low concentrations using a mos 2-based field-effect transistor,” RSC intraoperative detection of disseminated tumor cells in lymph nodes Advances, vol. 8, no. 15, pp. 7942–7948, 2018. with silicon nanowire field effect transistors,” ACS nano, vol. 10, no. 2, [190] S. Xu, J. Zhan, B. Man, S. Jiang, W. Yue, S. Gao, C. Guo, H. Liu, Z. Li, pp. 2357–2364, 2016. J. Wang et al., “Real-time reliable determination of binding kinetics of [210] X. Duan, Y. Li, N. K. Rajan, D. A. Routenberg, Y. Modis, and M. A. dna hybridization using a multi-channel graphene biosensor,” Nature Reed, “Quantification of the affinities and kinetics of protein inter- communications, vol. 8, p. 14902, 2017. actions using silicon nanowire biosensors,” Nature nanotechnology, [191] W. Fu, L. Feng, G. Panaitov, D. Kireev, D. Mayer, A. Offenhausser,¨ vol. 7, no. 6, p. 401, 2012. and H.-J. Krause, “Biosensing near the neutrality point of graphene,” [211] A. Gao, N. Lu, P. Dai, T. Li, H. Pei, X. Gao, Y. Gong, Y. Wang, and Science Advances, vol. 3, no. 10, p. e1701247, 2017. C. Fan, “Silicon-nanowire-based cmos-compatible field-effect transistor [192] J. Liu, J. Goud, P. M. Raj, M. Iyer, Z. Wang, and R. R. Tummala, nanosensors for ultrasensitive electrical detection of nucleic acids,” “Label-free protein detection by zno nanowire based bio-sensors,” in Nano letters, vol. 11, no. 9, pp. 3974–3978, 2011. Electronic Components and Technology Conference, 2007. ECTC’07. [212] F. Shen, M. Tan, Z. Wang, M. Yao, Z. Xu, Y. Wu, J. Wang, X. Guo, and Proceedings. 57th. IEEE, 2007, pp. 1971–1976. T. Zhu, “Integrating silicon nanowire field effect transistor, microflu- [193] F. Patolsky, G. Zheng, and C. M. Lieber, “Nanowire-based biosensors,” idics and air sampling techniques for real-time monitoring biological 2006. aerosols,” Environmental science & technology, vol. 45, no. 17, pp. [194] J. P. Kim, B. Y. Lee, J. Lee, S. Hong, and S. J. Sim, “Enhancement of 7473–7480, 2011. sensitivity and specificity by surface modification of carbon nanotubes [213] Y. Pei-Wen, H. Che-Wei, H. Yu-Jie, C. Min-Cheng, L. Hsin-Hao, in diagnosis of prostate cancer based on carbon nanotube field effect L. Shey-Shi, and L. Chih-Ting, “A device design of an integrated cmos transistors,” Biosensors and Bioelectronics, vol. 24, no. 11, pp. 3372– poly-silicon biosensor-on-chip to enhance performance of biomolecular 3378, 2009. analytes in serum samples,” Biosensors and Bioelectronics, vol. 61, pp. [195] B. Kim, H. S. Song, H. J. Jin, E. J. Park, S. H. Lee, B. Y. Lee, T. H. 112–118, 2014. Park, and S. Hong, “Highly selective and sensitive detection of neu- [214] K.-I. Chen, B.-R. Li, and Y.-T. Chen, “Silicon nanowire field- rotransmitters using receptor-modified single-walled carbon nanotube effect transistor-based biosensors for biomedical diagnosis and cellular sensors,” Nanotechnology, vol. 24, no. 28, p. 285501, 2013. recording investigation,” Nano today, vol. 6, no. 2, pp. 131–154, 2011. [196] Y.-S. Lo, D. H. Nam, H.-M. So, H. Chang, J.-J. Kim, Y. H. Kim, [215] B. Zhan, C. Li, J. Yang, G. Jenkins, W. Huang, and X. Dong, “Graphene and J.-O. Lee, “Oriented immobilization of antibody fragments on ni- field-effect transistor and its application for electronic sensing,” Small, decorated single-walled carbon nanotube devices,” ACS nano, vol. 3, vol. 10, no. 20, pp. 4042–4065, 2014. no. 11, pp. 3649–3655, 2009. [216] G.-J. Zhang and Y. Ning, “Silicon nanowire biosensor and its appli- [197] G.-J. Zhang, M. J. Huang, Z. H. H. Luo, G. K. I. Tay, E.-J. A. cations in disease diagnostics: a review,” Analytica Chimica Acta, vol. Lim, E. T. Liu, and J. S. Thomsen, “Highly sensitive and reversible 749, pp. 1–15, 2012. silicon nanowire biosensor to study nuclear hormone receptor protein [217] R. G. Hobbs, N. Petkov, and J. D. Holmes, “Semiconductor nanowire and response element dna interactions,” Biosensors and Bioelectronics, fabrication by bottom-up and top-down paradigms,” Chemistry of vol. 26, no. 2, pp. 365–370, 2010. Materials, vol. 24, no. 11, pp. 1975–1991, 2012. [198] Y. Ohno, K. Maehashi, and K. Matsumoto, “Label-free biosensors [218] D. Tran, T. Pham, B. Wolfrum, A. Offenhausser,¨ and B. Thierry, based on aptamer-modified graphene field-effect transistors,” Journal “Cmos-compatible silicon nanowire field-effect transistor biosen- of the American Chemical Society, vol. 132, no. 51, pp. 18 012–18 013, sor: Technology development toward commercialization,” Materials, 2010. vol. 11, no. 5, p. 785, 2018. [199] H. S. Song, H. J. Jin, S. R. Ahn, D. Kim, S. H. Lee, U.-K. Kim, [219] K. Maehashi and K. Matsumoto, “Label-free electrical detection using C. T. Simons, S. Hong, and T. H. Park, “Bioelectronic tongue using carbon nanotube-based biosensors,” Sensors, vol. 9, no. 7, pp. 5368– heterodimeric human taste receptor for the discrimination of sweeteners 5378, 2009. with human-like performance,” ACS nano, vol. 8, no. 10, pp. 9781– [220] S. Smart, A. Cassady, G. Lu, and D. Martin, “The biocompatibility of 9789, 2014. carbon nanotubes,” Carbon, vol. 44, no. 6, pp. 1034–1047, 2006. [200] C.-M. Lim, J. Y. Kwon, and W.-J. Cho, “Field-effect transistor biosen- [221] S. Kumar, R. Rani, N. Dilbaghi, K. Tankeshwar, and K.-H. Kim, sor platform fused with drosophila odorant-binding proteins for instant “Carbon nanotubes: a novel material for multifaceted applications in 37

human healthcare,” Chemical society reviews, vol. 46, no. 1, pp. 158– [242] H. Wang, Y. Wu, Z. Ni, and Z. Shen, “Electronic transport and layer 196, 2017. engineering in multilayer graphene structures,” Applied Physics Letters, [222] M. Sireesha, V. Jagadeesh Babu, A. S. Kranthi Kiran, and S. Ramakr- vol. 92, no. 5, p. 053504, 2008. ishna, “A review on carbon nanotubes in biosensor devices and their [243] J.-H. Chen, C. Jang, S. Xiao, M. Ishigami, and M. S. Fuhrer, “Intrinsic applications in medicine,” Nanocomposites, vol. 4, no. 2, pp. 36–57, and extrinsic performance limits of graphene devices on sio 2,” Nature 2018. nanotechnology, vol. 3, no. 4, p. 206, 2008. [223] S. J. Park, O. S. Kwon, S. H. Lee, H. S. Song, T. H. Park, and J. Jang, [244] K. I. Bolotin, K. Sikes, Z. Jiang, M. Klima, G. Fudenberg, J. Hone, “Ultrasensitive flexible graphene based field-effect transistor (fet)-type P. Kim, and H. Stormer, “Ultrahigh electron mobility in suspended bioelectronic nose,” Nano letters, vol. 12, no. 10, pp. 5082–5090, 2012. graphene,” Solid State Communications, vol. 146, no. 9-10, pp. 351– [224] J. Ping, R. Vishnubhotla, A. Vrudhula, and A. C. Johnson, “Scalable 355, 2008. production of high-sensitivity, label-free dna biosensors based on back- [245] A. C. Neto, F. Guinea, N. M. Peres, K. S. Novoselov, and A. K. Geim, gated graphene field effect transistors,” ACS nano, vol. 10, no. 9, pp. “The electronic properties of graphene,” Reviews of modern physics, 8700–8704, 2016. vol. 81, no. 1, p. 109, 2009. [225] N. Gao, T. Gao, X. Yang, X. Dai, W. Zhou, A. Zhang, and C. M. Lieber, [246] T. Enoki, Y. Kobayashi, and K.-I. Fukui, “Electronic structures of “Specific detection of biomolecules in physiological solutions using graphene edges and nanographene,” International Reviews in Physical graphene transistor biosensors,” Proceedings of the National Academy Chemistry, vol. 26, no. 4, pp. 609–645, 2007. of Sciences, vol. 113, no. 51, pp. 14 633–14 638, 2016. [247] V. Shenoy, C. Reddy, A. Ramasubramaniam, and Y. Zhang, “Edge- [226] A. K. Manoharan, S. Chinnathambi, R. Jayavel, and N. Hanagata, stress-induced warping of graphene sheets and nanoribbons,” Physical “Simplified detection of the hybridized dna using a graphene field effect review letters, vol. 101, no. 24, p. 245501, 2008. transistor,” Science and Technology of advanced MaTerialS, vol. 18, [248] A. K. Geim and K. S. Novoselov, “The rise of graphene,” Nature no. 1, pp. 43–50, 2017. Materials, vol. 6, no. 3, pp. 183–191, mar 2007. [Online]. Available: [227] G. Wu, Z. Dai, X. Tang, Z. Lin, P. K. Lo, M. Meyyappan, and http://www.nature.com/articles/nmat1849 K. W. C. Lai, “Graphene field-effect transistors for the sensitive [249] V. Geringer, M. Liebmann, T. Echtermeyer, S. Runte, M. Schmidt, and selective detection of escherichia coli using pyrene-tagged dna R. Ruckamp,¨ M. C. Lemme, and M. Morgenstern, “Intrinsic and aptamer,” Advanced healthcare materials, vol. 6, no. 19, p. 1700736, extrinsic corrugation of monolayer graphene deposited on sio 2,” 2017. Physical review letters, vol. 102, no. 7, p. 076102, 2009. [228] S. Viswanathan, T. N. Narayanan, K. Aran, K. D. Fink, J. Paredes, [250] L. Brey and J. Palacios, “Exchange-induced charge inhomogeneities in P. M. Ajayan, S. Filipek, P. Miszta, H. C. Tekin, F. Inci et al., rippled neutral graphene,” Physical Review B, vol. 77, no. 4, p. 041403, “Graphene–protein field effect biosensors: glucose sensing,” Materials 2008. Today, vol. 18, no. 9, pp. 513–522, 2015. [251] V. C. Tung, M. J. Allen, Y. Yang, and R. B. Kaner, “High-throughput [229] A. Hassibi, S. Zahedi, R. Navid, R. W. Dutton, and T. H. Lee, solution processing of large-scale graphene,” Nature nanotechnology, “Biological shot-noise and quantum-limited signal-to-noise ratio in vol. 4, no. 1, p. 25, 2009. affinity-based biosensors,” Journal of applied physics, vol. 97, no. 8, [252] Z.-S. Wu, W. Ren, L. Gao, J. Zhao, Z. Chen, B. Liu, D. Tang, B. Yu, p. 084701, 2005. C. Jiang, and H.-M. Cheng, “Synthesis of graphene sheets with high [230] A. M. Berezhkovskii and A. Szabo, “Effect of ligand diffusion on occu- electrical conductivity and good thermal stability by hydrogen arc pancy fluctuations of cell-surface receptors,” The Journal of chemical discharge exfoliation,” ACS nano, vol. 3, no. 2, pp. 411–417, 2009. physics, vol. 139, no. 12, p. 09B610 1, 2013. [253] Z. Ni, Y. Wang, T. Yu, Y. You, and Z. Shen, “Reduction of fermi ve- [231] M. Pierobon and I. F. Akyildiz, “A statistical-physical model of locity in folded graphene observed by resonance raman spectroscopy,” interference in diffusion-based molecular nanonetworks.” IEEE Trans. Physical Review B, vol. 77, no. 23, p. 235403, 2008. Communications, vol. 62, no. 6, pp. 2085–2095, 2014. [254] M. Moreno-Moreno, A. Castellanos-Gomez, G. Rubio-Bollinger, [232] ——, “Intersymbol and co-channel interference in diffusion-based J. Gomez-Herrero, and N. Agra¨ıt, “Ultralong natural graphene nanorib- molecular communication,” in Communications (ICC), 2012 IEEE bons and their electrical conductivity,” Small, vol. 5, no. 8, pp. 924–927, International Conference on. IEEE, 2012, pp. 6126–6131. 2009. [233] M. Deen, M. Shinwari, J. Ranuarez,´ and D. Landheer, “Noise consider- ations in field-effect biosensors,” Journal of applied physics, vol. 100, [255] F. Schedin, A. Geim, S. Morozov, E. Hill, P. Blake, M. Katsnelson, no. 7, p. 074703, 2006. and K. Novoselov, “Detection of individual gas molecules adsorbed on [234] N. K. Rajan, X. Duan, and M. A. Reed, “Performance limitations graphene,” Nature materials, vol. 6, no. 9, p. 652, 2007. for nanowire/nanoribbon biosensors,” Wiley Interdisciplinary Reviews: [256] N.-R. Kim, N. Farsad, C.-B. Chae, and A. W. Eckford, “A universal Nanomedicine and Nanobiotechnology, vol. 5, no. 6, pp. 629–645, channel model for molecular communication systems with metal- 2013. oxide detectors,” in Communications (ICC), 2015 IEEE International [235] Y. Kutovyi, I. Zadorozhnyi, H. Hlukhova, V. Handziuk, M. Petrychuk, Conference on. IEEE, 2015, pp. 1054–1059. A. Ivanchuk, and S. Vitusevich, “Origin of noise in liquid-gated si [257] B.-H. Koo, C. Lee, H. B. Yilmaz, N. Farsad, A. Eckford, and C.-B. nanowire troponin biosensors,” Nanotechnology, vol. 29, no. 17, p. Chae, “Molecular mimo: From theory to prototype,” IEEE Journal on 175202, 2018. Selected Areas in Communications, vol. 34, no. 3, pp. 600–614, 2016. [236] M. Khalid, O. Amin, S. Ahmed, and M.-S. Alouini, “System [258] W. Wicke, A. Ahmadzadeh, V. Jamali, H. Unterweger, C. Alexiou, and Modeling of Virus Transmission and Detection in Molecular R. Schober, “Magnetic nanoparticle based molecular communication in Communication Channels,” in 2018 IEEE International Conference microfluidic environments,” arXiv preprint arXiv:1808.05147, 2018. on Communications (ICC). IEEE, may 2018, pp. 1–6. [Online]. [259] M. Pierobon, I. F. Akyildiz et al., “Diffusion-based noise analysis Available: https://ieeexplore.ieee.org/document/8422665/ for molecular communication in nanonetworks,” IEEE Transactions on [237] M. Pumera, “Graphene in biosensing,” Materials today, vol. 14, no. Signal Processing, vol. 59, no. 6, pp. 2532–2547, 2011. 7-8, pp. 308–315, 2011. [260] H. B. Yilmaz, A. C. Heren, T. Tugcu, and C. B. Chae, “Three- [238] K. S. Novoselov, A. K. Geim, S. Morozov, D. Jiang, M. Katsnelson, dimensional channel characteristics for molecular communications with I. Grigorieva, S. Dubonos, and A. Firsov, “Two-dimensional gas of an absorbing receiver,” IEEE Communications Letters, vol. 18, no. 6, massless dirac fermions in graphene,” nature, vol. 438, no. 7065, p. pp. 929–932, 2014. 197, 2005. [261] A. Noel, K. C. Cheung, and R. Schober, “Improving receiver perfor- [239] E. V. Castro, K. Novoselov, S. Morozov, N. Peres, J. L. Dos Santos, mance of diffusive molecular communication with enzymes,” IEEE J. Nilsson, F. Guinea, A. Geim, and A. C. Neto, “Biased bilayer Transactions on NanoBioscience, vol. 13, no. 1, pp. 31–43, 2014. graphene: semiconductor with a gap tunable by the electric field effect,” [262] ——, “A unifying model for external noise sources and isi in diffu- Physical review letters, vol. 99, no. 21, p. 216802, 2007. sive molecular communication,” IEEE Journal on Selected Areas in [240] M. Craciun, S. Russo, M. Yamamoto, J. B. Oostinga, A. Morpurgo, Communications, vol. 32, no. 12, pp. 2330–2343, 2014. and S. Tarucha, “Trilayer graphene is a semimetal with a gate-tunable [263] V. Jamali, A. Ahmadzadeh, and R. Schober, “On the design of matched band overlap,” Nature nanotechnology, vol. 4, no. 6, p. 383, 2009. filters for molecule counting receivers,” IEEE Commun. Lett, vol. 21, [241] H. Hibino, H. Kageshima, M. Kotsugi, F. Maeda, F.-Z. Guo, and no. 8, pp. 1711–1714, 2017. Y. Watanabe, “Dependence of electronic properties of epitaxial few- [264] A. Shahbazi and A. Jamshidi, “Adaptive weighted signal detection layer graphene on the number of layers investigated by photoelectron for nanoscale molecular communications,” in The International Zurich emission microscopy,” Physical Review B, vol. 79, no. 12, p. 125437, Seminar on Information and Communication (IZS 2018) Proceedings. 2009. ETH Zurich, 2018, pp. 149–152. 38

[265] P. He, Y. Mao, Q. Liu, and K. Yang, “Improving reliability performance [288] M. Kuscu and O. B. Akan, “Modeling convection-diffusion-reaction of diffusion-based molecular communication with adaptive threshold systems for microfluidic molecular communications with surface-based variation algorithm,” International Journal of Communication Systems, receivers in internet of bio-nano things,” PloS One, vol. 13, no. 2, p. vol. 29, no. 18, pp. 2669–2680, 2016. e0192202, 2018. [266] G. H. Alshammri, M. S. Alzaidi, W. K. Ahmed, and V. B. [289] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “Strength-based Lawrence, “Low-complexity memory-assisted adaptive-threshold de- optimum signal detection in concentration-encoded pulse-transmitted tection scheme for on-off-keying diffusion-based molecular commu- ook molecular communication with stochastic ligand-receptor binding,” nications,” in Sarnoff Symposium, 2017 IEEE 38th. IEEE, 2017, pp. Simulation Modelling Practice and Theory, vol. 42, pp. 189–209, 2014. 1–6. [290] C. T. Chou, “A markovian approach to the optimal demodulation of [267] M. Damrath and P. A. Hoeher, “Low-complexity adaptive thresh- diffusion-based molecular communication networks,” IEEE Transac- old detection for molecular communication,” IEEE Transactions on tions on Communications, vol. 63, no. 10, pp. 3728–3743, 2015. Nanobioscience, vol. 15, no. 3, pp. 200–208, 2016. [291] ——, “Maximum a-posteriori decoding for diffusion-based molecular [268] H. ShahMohammadian, G. G. Messier, and S. Magierowski, “Opti- communication using analog filters,” IEEE Transactions on Nanotech- mum receiver for molecule shift keying modulation in diffusion-based nology, vol. 14, no. 6, pp. 1054–1067, 2015. molecular communication channels,” Nano Communication Networks, [292] G. Muzio, M. Kuscu, and O. B. Akan, “Selective signal detection vol. 3, no. 3, pp. 183–195, 2012. with ligand receptors under interference in molecular communications,” [269] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “A comprehensive in 2018 IEEE 19th International Workshop on Signal Processing analysis of strength-based optimum signal detection in concentration- Advances in Wireless Communications (SPAWC). IEEE, 2018, pp. encoded molecular communication with spike transmission,” IEEE 1–5. Transactions on Nanobioscience, vol. 14, no. 1, pp. 67–83, 2015. [293] I. F. Akyildiz, “Nanonetworks: A new frontier in communications,” [270] ——, “A comprehensive study of sampling-based optimum signal in Proceedings of the 18th annual international conference on Mobile detection in concentration-encoded molecular communication,” IEEE computing and networking. ACM, 2012, pp. 1–2. Transactions on Nanobioscience, vol. 13, no. 3, pp. 208–222, 2014. [271] H. Yan, G. Chang, Z. Ma, and L. Lin, “Derivative-based signal [294] G. Zheng, X. P. Gao, and C. M. Lieber, “Frequency domain detection of Nano letters detection for high data rate molecular communication system,” IEEE biomolecules using silicon nanowire biosensors,” , vol. 10, Communications Letters, vol. 22, no. 9, pp. 1782–1785, 2018. no. 8, pp. 3179–3183, 2010. [272] B. Li, M. Sun, S. Wang, W. Guo, and C. Zhao, “Low-complexity non- [295] G. S. Kulkarni and Z. Zhong, “Detection beyond the debye screening coherent signal detection for nanoscale molecular communications,” length in a high-frequency nanoelectronic biosensor,” Nano letters, IEEE Transactions on Nanobioscience, vol. 15, no. 1, pp. 3–10, 2016. vol. 12, no. 2, pp. 719–723, 2012. [273] ——, “Local convexity inspired low-complexity noncoherent signal [296] W. Fu, L. Jiang, E. P. van Geest, L. M. Lima, and G. F. Schneider, detector for nanoscale molecular communications,” IEEE Transactions “Sensing at the surface of graphene field-effect transistors,” Advanced on Communications, vol. 64, no. 5, pp. 2079–2091, 2016. Materials, vol. 29, no. 6, p. 1603610, 2017. [274] V. Jamali, N. Farsad, R. Schober, and A. Goldsmith, “Non-coherent [297] B. Atakan and O. B. Akan, “On channel capacity and error compensa- detection for diffusive molecular communication systems,” IEEE Trans- tion in molecular communication,” in Transactions on computational actions on Communications, 2018. systems biology X. Springer, 2008, pp. 59–80. [275] V. Jamali, A. Ahmadzadeh, N. Farsad, and R. Schober, “Constant- [298] N. Farsad, N.-R. Kim, A. W. Eckford, and C.-B. Chae, “Channel and composition codes for maximum likelihood detection without csi in noise models for nonlinear molecular communication systems,” IEEE diffusive molecular communications,” IEEE Transactions on Commu- Journal on Selected Areas in Communications, vol. 32, no. 12, pp. nications, vol. 66, no. 5, pp. 1981–1995, 2018. 2392–2401, 2014. [276] A. Noel and A. W. Eckford, “Asynchronous peak detection for demod- [299] S. Abadal and I. F. Akyildiz, “Bio-inspired synchronization for ulation in molecular communication,” in Communications (ICC), 2017 nanocommunication networks,” in Global Telecommunications Confer- IEEE International Conference on. IEEE, 2017, pp. 1–6. ence (GLOBECOM 2011), 2011 IEEE. IEEE, 2011, pp. 1–5. [277] G. Chang, L. Lin, and H. Yan, “Adaptive detection and isi mitigation [300] H. ShahMohammadian, G. G. Messier, and S. Magierowski, “Blind for mobile molecular communication,” IEEE Transactions on Nanobio- synchronization in diffusion-based molecular communication chan- science, vol. 17, no. 1, pp. 21–35, 2018. nels,” IEEE communications letters, vol. 17, no. 11, pp. 2156–2159, [278] A. Ahmadzadeh, V. Jamali, and R. Schober, “Stochastic channel 2013. modeling for diffusive mobile molecular communication systems,” [301] L. Lin, C. Yang, M. Ma, S. Ma, and H. Yan, “A clock synchronization arXiv preprint arXiv:1709.06785, 2017. method for molecular nanomachines in bionanosensor networks,” IEEE [279] Y. Murin, N. Farsad, M. Chowdhury, and A. Goldsmith, “Time-slotted Sensors Journal, vol. 16, no. 19, pp. 7194–7203, 2016. transmission over molecular timing channels,” Nano Communication [302] V. Jamali, A. Ahmadzadeh, and R. Schober, “Symbol synchronization Networks, vol. 12, pp. 12–24, 2017. for diffusive molecular communication systems,” in 2017 IEEE Inter- [280] ——, “Optimal detection for diffusion-based molecular timing chan- national Conference on Communications (ICC). IEEE, 2017, pp. 1–7. arXiv preprint arXiv:1709.02977 nels,” , 2017. [303] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “Concentration- [281] B. C. Akdeniz, A. E. Pusane, and T. Tugcu, “Optimal reception delay encoded subdiffusive molecular communication: Theory, channel char- IEEE Communications in diffusion-based molecular communication,” acteristics, and optimum signal detection.” IEEE transactions on Letters , vol. 22, no. 1, pp. 57–60, 2018. nanobioscience, vol. 15, no. 6, pp. 533–548, 2016. [282] T. Mora, “Physical limit to concentration sensing amid spurious lig- [304] S. M. Mustam, S. K. Syed Yusof, and S. Nejatian, “Multilayer ands,” Physical review letters, vol. 115, no. 3, p. 038102, 2015. diffusion-based molecular communication,” Transactions on Emerging [283] M. Kuscu and O. B. Akan, “Maximum likelihood detection with ligand Telecommunications Technologies, vol. 28, no. 1, p. e2935, 2017. receptors for diffusion-based molecular communications in internet of bio-nano things,” IEEE transactions on nanobioscience, vol. 17, no. 1, [305] V. Jamali, A. Ahmadzadeh, C. Jardin, H. Sticht, and R. Schober, pp. 44–54, 2018. “Channel estimation for diffusive molecular communications,” IEEE [284] A. Ahmadzadeh, H. Arjmandi, A. Burkovski, and R. Schober, “Com- Transactions on Communications, vol. 64, no. 10, pp. 4238–4252, prehensive reactive receiver modeling for diffusive molecular commu- 2016. nication systems: Reversible binding, molecule degradation, and finite [306] A. Noel, K. C. Cheung, and R. Schober, “Joint channel parameter number of receptors,” IEEE transactions on nanobioscience, vol. 15, estimation via diffusive molecular communication,” IEEE Transactions no. 7, pp. 713–727, 2016. on Molecular, Biological and Multi-Scale Communications, vol. 1, [285] H. ShahMohammadian, G. G. Messier, and S. Magierowski, “Mod- no. 1, pp. 4–17, 2015. elling the reception process in diffusion-based molecular communi- [307] V. Singh, M. Tchernookov, and I. Nemenman, “Effects of receptor cation channels,” in Communications Workshops (ICC), 2013 IEEE correlations on molecular information transmission,” Physical Review International Conference on. IEEE, 2013, pp. 782–786. E, vol. 94, no. 2, p. 022425, 2016. [286] C. T. Chou, “Extended master equation models for molecular com- [308] D. Bray, M. D. Levin, and C. J. Morton-Firth, “Receptor clustering munication networks,” IEEE Transactions on Nanobioscience, vol. 12, as a cellular mechanism to control sensitivity,” Nature, vol. 393, no. no. 2, pp. 79–92, 2013. 6680, p. 85, 1998. [287] A. Einolghozati, M. Sardari, and F. Fekri, “Capacity of diffusion- [309] A. Mugler, F. Tostevin, and P. R. Ten Wolde, “Spatial partitioning based molecular communication with ligand receptors,” in Information improves the reliability of biochemical signaling,” Proceedings of the Theory Workshop (ITW), 2011 IEEE. IEEE, 2011, pp. 85–89. National Academy of Sciences, p. 201218301, 2013. 39

[310] K. Kaizu, W. De Ronde, J. Paijmans, K. Takahashi, F. Tostevin, and P. R. Ten Wolde, “The berg-purcell limit revisited,” Biophysical journal, vol. 106, no. 4, pp. 976–985, 2014. [311] C. T. Chou, “Molecular circuit for approximate maximum a posteriori demodulation of concentration modulated signals,” in Proceedings of the 5th ACM International Conference on Nanoscale Computing and Communication. ACM, 2018, p. 11. [312] ——, “Chemical reaction networks for maximum likelihood estimation of the concentration of signalling molecules,” in Proceedings of the 4th ACM International Conference on Nanoscale Computing and Communication. ACM, 2017, p. 4. [313] J.-B. Lalanne and P. Franc¸ois, “Chemodetection in fluctuating environ- ments: Receptor coupling, buffering, and antagonism,” Proceedings of the National Academy of Sciences, vol. 112, no. 6, pp. 1898–1903, 2015. [314] P. Franc¸ois and G. Altan-Bonnet, “The case for absolute ligand dis- crimination: modeling information processing and decision by immune t cells,” Journal of Statistical Physics, vol. 162, no. 5, pp. 1130–1152, 2016.