For Future Electronics
Materials for future electronics
by Kwan S. Kwok
The Defense Advanced Research Projects Agency A previous article in Materials Today1described (DARPA) is continuing its far-reaching vision, from DARPA’s Moletronics Program and its intense interest the invention of stealth fighters to creating a new in developing molecular-based electronic materials for use in both memory architectures and circuit class of computers for the future. Future computers elements2-9. This article is a continuation of that must be small, lightweight, low-power, and have previous one and provides an update on the program properties that will eventually yield computational and its latest accomplishments. capabilities reaching or exceeding those of the human brain. The Moletronics Program at DARPA was Memories of the future created first to optimize size, weight, power, and The primary objective of DARPA’s Moletronics Program is to deliver a prototype 16 kbit nanocomputer memory by early parallelism, and then to design the required elements 2005. This prototype will be integrated on the molecular for computation around basic building units. These scale and have dimensions of ~10 µm x 10 µm, which is will one day revolutionize the current semiconductor approximately the footprint of a human cell. As a result, the 2 industry by providing Si devices with additional target device density for the nanomemory is 100 Gbit/cm . At this level of integration, the prototype will be ten times functions that will create new applications. more dense than the dynamic random access memory (DRAM) that the Si industry projects it will deliver at the end of its roadmap in 2016. Additional objectives for the Moletronics Program’s nanomemory are that it should be nonvolatile and low-power, as well as defect- and fault- tolerant. It should also have an input/output interface with microelectronic systems and be fabricated using new techniques for hierarchical self-assembly, which will eventually provide a path to cost-effective mass manufacturing of the system. Several complementary technical approaches are being Moletronics Program, applied successfully by the Moletronics Program to Microsystems Technology Office, Defense Advanced Research Projects Agency, demonstrate molecular-scale electronic nanomemory arrays. 3701 North Fairfax Drive, These include: Arlington, VA 22203-1714, USA E-mail: [email protected] • Switches made from electrically active molecules
20 December 2003 ISSN:1369 7021 © Elsevier Ltd 2003 PROGRESS REPORT
sandwiched between imprinted metal nanowires at the intersections of a nanowire crossbar array; • Arrays of nanoscale diodes and transistors formed at the intersections of doped, self-assembled semiconductor nanowires; and • Stacks of electrically active porphyrin molecules, which are used to enhance the density of conventional Si CMOS (complementary metal-oxide semiconductor) DRAM. Key inventions and innovations of the program have included: novel molecular switches and wires, molecular-scale diodes and transistors built from nanowires, plus techniques for the Fig. 1 An atomic force micrograph of an 8 x 8 crossbar circuit. precise, mass fabrication and assembly/imprinting of very large numbers of molecule-sized nanowires, new molecular The basic element in the circuit is the Pt/molecule/Ti electronic devices, and system simulations that have been junction formed at each cross-point, which acts as a written especially for the Moletronics Program. These are reversible and nonvolatile switch; 64 such switches are being used extensively to guide the research and to ensure connected to form an 8 x 8 crossbar circuit within a 1 µm2 the functionality of proposed designs and architectures in area. To demonstrate demultiplexer/multiplexer functionality advance, thereby shortening the entire development cycle. integrated with memory, a defect-free 8 x 8 crossbar was configured into a 4 x 4 memory and two 4 x 4 decoders by Molecular electronic circuits setting the resistances at specific cross-points. In the Hewlett-Packard Laboratories has developed a process to demultiplexer/multiplexer logic circuits, various address codes fabricate crossbar molecular circuits with the highest-density were used to select and transmit signals from multiple input electronically addressable memory reported to date. The lab wires to a single output wire (and/or from a single input wire demonstration circuit – a 64 bit memory using molecules as to one of several output wires) to read the resistances in the switches – occupies an area of 1 µm2,as shown in Fig. 1. The memory. bit density of the circuit is more than ten times greater than In addition to the development of high-density molecular today’s Si memory chips. It also combines both memory and electronic circuits, which form critical building blocks for a logic for the first time, using rewritable, nonvolatile molecular computing system, molecular memory devices molecular-switch devices. The circuits were fabricated using were incorporated onto CMOS to form a hybrid system. This nano-imprint lithography. approach provides an early opportunity for the inclusion of In the circuits, the bottom nanowire electrodes are defined molecular devices in conventional Si-based memory products. by imprint lithography. A molecular monolayer of rotaxane is deposited using the Langmuir Blodgett method9-11(Fig. 2). Hybrid memory devices Fabrication of the top electrodes begins with the blanket Researchers at the University of California-Riverside, North evaporation of a 7.5 nm Ti protective layer, which minimizes Carolina State University (NCSU), and ZettaCore, Inc. have subsequent disruption of the molecular monolayer and also functions as a direct electrical contact to the molecules. Patterned top electrodes of 5 nm Ti and 10 nm Pt are then fabricated by the same process, using the same imprint mold oriented perpendicularly to the bottom electrodes. Finally, reactive ion etching (RIE) is used to remove the Ti protective layer, leaving the Pt electrodes intact. The molecules and Ti layer under the Pt top electrode are protected. After RIE, the crossbar circuits with the molecular monolayer sandwiched between bottom and top nanowires remain. Fig. 2 A schematic diagram showing the circuit structure.
December 2003 21 PROGRESS REPORT
While the hybrid device represents an early opportunity for the inclusion of molecular devices into a computing product, a random assembly approach based on a cell containing nano-sized components, or NanoCell, represents a bold new way to achieve manufacturing and assembly. The NanoCell concept combines present Si-based technology and that based purely on molecular switches and wires. If successful, the NanoCell random assembly approach will revolutionize the way in which computers are built and Fig. 3 A porphyrin memory molecule. circuits are reconfigured and perform computation.
recently demonstrated the first fully-functional hybrid NanoCells for molecular computing CMOS/molecular memory device12. The development of Nanoelectronic architectures could prove themselves capable hybrid devices is an important first step in establishing a of complementing traditional solid-state devices13-15. Most transitional technology that will lead to a class of fully proposed architectures are dependent on precise order and molecular computers. building devices with exact arrays of nanostructures, which In the hybrid device, porphyrin molecules, as shown in are painstakingly interfaced with microstructure16-20. Fig. 3, are attached to a lithographically fabricated Si Conversely, a team from Rice, NCSU, Yale, South Carolina platform in an array of memory cells, as shown in Fig. 4. Bits (USC) and Penn State Universities, as well as Motorola Corp., of information are stored in the discrete redox states of the has been developing the NanoCell concept for nanoelectronic porphyrin molecules. The porphyrins are particularly assemblies. The NanoCell approach13-21is not dependent attractive candidates for memory applications because they upon placing molecules or nanowires in precise orientations have the ability to store multiple bits of information (certain or locations. The internal portions are, for the most part, porphyrin-based molecules are capable of storing three bits disordered and there is no need to precisely locate any of the of information). This will greatly reduce the complexity and switching elements. The nano-sized switches are added in increase the storage capacity of the memory chip. In the abundance, yet only a small percentage are needed to prototypical CMOS/molecular device (Fig. 4), the memory array is fully integrated with on-chip sense amplifiers, which are fabricated by conventional methods. These sense amplifiers have been designed to read multiple bits.
Fig. 5 A scanning electron micrograph of the NanoCell after assembly of the Au nanowires and molecule 134. The upper image shows the five juxtaposed pairs of fabricated leads across the NanoCell; some Au nanowires are barely visible on the internal rectangle of the discontinuous Au film. The lower image is the NanoCell's central portion, at a higher magnification, showing the disordered discontinuous Au film with an attached Au nanowire that is affixed via the OPE-dithiol (not observable) derived from molecule 1. Fig. 4 A prototype hybrid CMOS/molecular memory chip. (Reprinted with permission from34. © 2003 American Chemical Society.)
22 December 2003 PROGRESS REPORT
Molecular programmability Molecular electronics can be developed if we are able to program a quasi-random arrangement of molecules or a field- programmable molecular random array23-25. The moletronics group at USC has demonstrated that even two highly nonlinear devices connected in series yield a multi-valued response. Furthermore, it is this multi-valued response that allows the reprogramming of a molecular device. This is possible because of the USC team’s recent progress in the theoretical bottom-up approach to the design of molecular electronic circuits. The team has implemented a precise Fig. 6 The I-V characteristics of the NanoCell before (a-c) and after (d-f) a voltage set- pulse of -8 V for 2 s at 297 K. The initial high-σ state (0-state) is represented by curves procedure to determine the current-voltage (I-V) a, b, and c, with the low-σ state (1-state) by curves d, e, and f, which are the 1st, 2nd and 3rd scans before and after the set-pulse, respectively. The inset shows curves d-f in the characteristics of integrated molecular circuits, where the microamp range. There is a 400:1 ratio in 0-state to 1-state current levels between the systems are analyzed at several levels on the basis of theory. high- and low-σ states, recorded at -2 V for this device. This allows the interfacing of systems from fractions of assemble in an orientation for switching. The lithographic nanometers to several micrometers, i.e. the nano-micro challenges of the input/output structures in the NanoCell interface. Extensive studies have been performed on become far less exacting and the fault tolerance is enormous; molecular circuits in which the geometry and nature of the however, programming is more challenging. contacts were critically analyzed. This work describes the first example in which a NanoCell The programmability feature is the most important is actually assembled (Fig. 5). Remarkably, the NanoCell concept in molecular electronics since this compensates for exhibits reproducible, room temperature negative differential the lack of addressability at the atomic scale. The USC resistance (NDR)22switching behavior with excellent peak- procedure allows us to find solutions in typical configurations to-valley (PVR) ratios, peak currents in the milliamp range, when molecular devices are self-assembled. Fig. 7 shows how and reprogrammable memory states that are stable for more the single molecule nonlinear I-V characteristics were than a week, even upon exposure to air, with substantial calculated using ab initio techniques. Consequently, the I-V 0:1 bit level ratios of 102-103, sometimes as high as 106 characteristics of a series configuration of two identical (Fig. 6). Metallic nanofilaments have been shown to be the devices is obtained (Fig. 7, bottom left). The I-Vtrace clearly chief mechanism of action in these present-generation shows multi-valued features. For instance, two current values embodiments. can be obtained for any applied voltage in the range A NanoCell assembled with disordered arrays of molecules 2.5-3.5 V. This means that, for a given bias voltage in that and Au nanowires has been fabricated, which operates via the nanofilaments that form during testing. The NanoCell exhibits reproducible switching behavior and two types of memory effects: one is a destructive read; the other is a nondestructive read. This first demonstration of a disordered ensemble for high-yielding switching and memory promises to yield a disordered programmable array for complex device functionality13. An important component of the NanoCell approach is a theoretical understanding and the modeling of the devices in their most fundamental states. Modeling and simulating tools developed for circuit analyses during the design and construction of the NanoCell have been further refined to Fig. 7 From the single-molecule nonlinear I-V characteristics calculated using the most precise ab initio techniques (top left), the I-V characteristics of a series configuration of address an area of importance – molecular programmability. two identical devices are obtained (bottom left).
December 2003 23 PROGRESS REPORT
modeling tools and an improvement in the fundamental understanding of the behavior of molecular-scale devices as individual molecules. This will provide further insight into conductance behaviors in individual molecules.
Conduction in individual molecules The ability to design molecular devices that exhibit specific electronic properties is predicated on having a firm Fig. 8 Top: an illustration of a molecular junction. Bottom left: a field-emission SEM image of a junction prior to electromigration. Bottom right: I-V characteristics of an understanding of the fundamental mechanisms responsible electromigrated junction. for conduction in individual and small bundles of molecules. range, two possible states can be obtained, distinguished by Researchers at Penn State University have measured the
different values of current. Thus, external impulses may electrical transport properties of single NO2-substituted trigger sudden changes of state, allowing the series circuit to phenylene-ethynylene (NOPE) molecular wires that span a be programmed to follow a specific path in the I-V plane. nanoscale Au gap to elucidate the role of thermal effects on Notice that, in this particular case, the two active molecules conduction through this molecule26-28. Inelastic tunneling are separated by an Au cluster. However, this is not limited spectroscopy (IETS) of such junctions reveals clear peaks in to one Au cluster, but could simply be produced by a series of d2I/dV2 spectra that are characteristic of the molecular two active rings. This explains why two peaks are sometimes vibrations of this molecule. This provides evidence that the obtained in the I-V characteristics. It also explains why one I- measured I-V behavior is governed by properties that are V curve is obtained in one sweep and a different one in specific to this molecule (Figs. 8 and 9). Temperature- another. In the case of more than three devices, the richness dependent I-V characteristics show a clear transition from of possibilities increases to allow configurations with gain temperature-independent tunneling at low temperatures to and memory characteristics (Fig. 7, bottom right). Because of activated hopping at higher temperatures. These results the multi-valued nature of nonlinear devices, this is one of confirm theoretical predictions of a thermally mediated the most important properties through which the transition in the molecular conduction mechanism. characteristics can be programmed externally. These features Agreement between theoretical models and experimental are fully exploited in the construction of a NanoCell at the observations provides pathways to build circuits with larger end of the arrows (Fig. 7, top center), which can be architectures. Another new approach supported by the interconnected to other NanoCells by standard lithographic Moletronics Program is the use of biologically-derived techniques. nanostructures to build electronic circuits on the nanoscale. The validation of modeling and simulation tools is usually accomplished by comparing measured data to those Nanoelectronics on a bio-scaffold calculated from models. An important component of the One of the most important challenges in the fabrication of a Moletronics Program is the validation of simulation and nanoelectronic device is the creation of a nano-sized scaffold
Fig. 9 Left: IETS of a typical molecular junction with peak assignments corresponding to NOPE. Right: Arrhenius plot of ln(I) vs 1000/T at bias voltages of 0.1-1 V (lowest curve to highest).
24 December 2003 PROGRESS REPORT
elements is seen after the attachment of the molecules. Fig. 12 shows a model of the multiple pathways for the passage of current across the VNB30. The use of a VNB as a nano-sized scaffold for molecular circuits and the NanoCell for the assembly of complex electronic devices are just two of many components in the Moletronics Program that leverage tools useful for other applications in molecular electronics.
Fig. 10 A schematic diagram of a VNB. Nanowire heterostructures and lasers Charles Lieber’s group at Harvard University has made onto which a desired nanoelectronic circuit can be wired. revolutionary advances in the growth of nano/molecular- This requires engineering of the scaffold on molecular length scale building blocks, with applications ranging from high- scales (2-3 nm). The moletronics team at the US Naval performance molecular electronics and computing to Research Laboratory (NRL), in collaboration with scientists nanophotonics, including the first nanoscale electrically from the Scripps Institute, Geo-Centers, NCSU, and the pumped laser31-33. University of North Texas, have been working on the concept The researchers have developed a general approach for the of using a biologically derived virus nanoparticle or virus synthesis of semiconductor nanowire superlattices from III-V nanoblock (VNB) as the scaffold on which to build a and IV materials, in which the composition and/or doping are nanoelectronic circuit. varied along the axis of the nanowire on a molecular scale31. The various steps in the fabrication of a circuit on a VNB The Harvard group has demonstrated the growth of involve the insertion of point-specific mutations on the VNB GaAs/GaP compositionally-modulated superlattices, n-Si/p-Si using molecular biology, the attachment of Au nanoparticles and n-InP/p-InP modulation-doped nanowires containing up to the modified locations, and, finally, the attachment of to 21 layers. Moreover, they have shown that these new molecular wires to bridge the Au nanoparticles29. Fig. 10 systems can function as molecular electronic devices, nano- shows a schematic diagram of the molecular electronic scale bar codes, and polarized nano-sized light-emitting network wired onto the Au nanoparticles, which are attached diodes (Fig. 13). to specific locations on the VNB. Lieber has also invented an approach for the growth of The electronic properties of the molecular electronic core- and multi-shell nanowire heterostructures that is circuit on the VNB have been determined by scanning applicable to many nanoscale materials32. He has tunneling microscopy (STM). Fig. 11 shows the I-V demonstrated the growth of homoepitaxial and characteristics of the VNB before and after the attachment of heteroepitaxial Si and SiGe core-shell nanostructures, and the molecular electronic elements. The I-V curves show that used these novel nanomaterials to make a high-performance, there is no current without the molecular network, while coaxially gated field-effect transistor (Fig. 14). More typical I-V traces characteristic of a network of conductive generally, the research described here opens up many
Fig. 11 I-V characteristics of a VNB with and without circuit molecules. Fig. 12: A model of a VNB nanoelectronic circuit with multiple pathways for current flow.
December 2003 25 PROGRESS REPORT
Conventional miniaturized semiconductor laser diodes used for optical telecoms and optical storage (such as compact discs) are cumbersome. This means that optoelectronic technology, for example in the form of circuits for converting electronic signals to optical pulses for fiber- optic transmission lines, suffers because the interface between microelectronics and photonic devices is problematic. It would be far better if the lasers could be mounted directly on chips. Some degree of integration has been made possible by microlasers, such as vertical-cavity surface-emitting lasers, in which a quantum well laser cavity is sandwiched between multilayer dielectric mirrors made
Fig. 13 Images of emission from nanoscale GaAs regions in GaP/GaAs nanowire from thin films of semiconductors. However, these devices superlattices containing 7, 11, and 21 layers31. are still gargantuan – several microns across – compared with (Reprinted by permission from31. © Macmillan Publishers Ltd.) the nanowire lasers that Lieber has developed. In effect, the opportunities in molecular electronics and nanophotonics, new devices are miniaturized versions of fiber lasers, in which ranging from relatively simple nanoscale emitters and a conventional hair-thin optical fiber supplies the laser cavity. complementary logic to complex periodic superlattices, which Fiber lasers are widespread in telecoms and medical could enable applications such as nanowire injection lasers technologies. In the nanowire laser, the wire acts as a and ‘engineered’ one-dimensional electron waveguides. waveguide for the stimulated emission. Lieber’s group has Significantly, this work has also led to the creation of shrunk the fiber laser at least a thousand-fold. In addition, he electrically driven lasers from individual semiconducting has made nanowires of other semiconductors with various wires33. The nanowire lasers consist of single wires of CdS, band gaps, such as GaN and InP. Thus, it should be possible to which emits green light at wavelengths of about 495-500 nm, make nanoscale lasers that span the spectrum from sandwiched between p-type Si and a metal electrode. The ultraviolet to infrared (Fig. 16). These devices might be small nanowires act as n-type semiconductors (electron enough to allow chip-scale integration of optoelectronics and transporters), while the Si is doped so that it is p-type (hole photonics at a truly unprecedented level. transporting). The interface of the two forms a p-n junction, The applications of nanowire lasers may not be limited to across which charge can be injected into the nanowire. There, optical telecoms and IT. Lieber suggests that it might be the recombination of charge carriers causes the emission of possible to use nanowire lasers to perform surgery with blue-green light. Lasing turns on above a threshold current of unprecedented precision, or as optical probes for chemical ~200 µA, where a broad spontaneous-emission peak collapses and biological sensors (the nanowires may penetrate through into very sharp peaks that are the typical fingerprint of lasing cell walls without causing much damage). Furthermore, they (Fig. 15). could be used for very high-resolution lithography, making
Fig. 14 Left: a transmission electron micrograph of epitaxial single-crystal Ge/Si core/shell nanowire heterostructures. Right: a schematic diagram of a more complex coaxial core/ Fig. 15 An image of a CdS nanowire injection laser, with light emitted from the nanowire shell nanowire transistor32. (Reprinted by permission from32. © Macmillan Publishers Ltd.) end at the center-right of the image.
26 December 2003 PROGRESS REPORT
DRAM that the Si industry has as its goal at the end of its roadmap in 2016. As a result, no longer will the world speak of electronic microsystems, but only of ‘nanosystems’. To put this in perspective, the prototype 16 kbit nanomemory system will be no larger than a human cell. Furthermore, it will be a nonvolatile molecular memory with the ability to retain its contents without power, thus also reducing the size requirements for supporting systems and Fig. 16 An illustration of an integrated multicolor nanowire laser array formed by the assembly of different composition nanowires into an injection laser structure33. structures. Based on these initial successes in molecular-scale electronics, DARPA is expanding moletronics development use of the optical near-field to beat the wavelength beyond nanomemory to molecular nanoprocessing systems resolution limit. and nanosensing systems. The moletronics technology is taking device scaling beyond the Si CMOS limit – that is, Electronics beyond the conventional ‘beyond the wall’ – for a range of processing and sensing The DARPA Moletronics Program has already used molecular applications. We anticipate that these revolutionary switches to produce a prototype 64 bit memory with a innovations will have a sweeping impact on defense systems, density of 6.4 Gigabit/cm2, which is ten times denser than increasing capabilities and reducing the weight and form currently available Si DRAM. The Moletronics Program will factor for platforms that depend on electronics: and these continue to push electronics technology further – all the way days, of course, that means all platforms. NT to the molecular and quantum domains – to create a new breed of radio frequency and mixed-signal electronics. By Acknowledgments The author would like to acknowledge and express much appreciation for the many early 2005, the program will have delivered a prototype valuable discussions and informative correspondence with the investigators participating 16 kbit memory with a density of 100 Gigabit/cm2. This in the DARPA Moletronics Program, all of which contributed greatly to this paper. This work was sponsored by the US Defense Advanced Research Projects Agency. 2004-2005 prototype will be ten times denser than the
REFERENCES 1. Kwok, K. S., and Ellenbogen, J. C., Materials Today (2002), 5 (7), 28 18. Goldstein, S. C., and Budiu, M., Proc. ISCA (2001), 178 2. Ellenbogen, J. C., and Love, J. C., Proc. IEEE (2000) 88, 386, and references 19. Melosh, N. A., et al., Science (2003) 300, 112 therein 20. Huang, Y., et al., Science (2001) 294, 1313 3. Joachim, C., et al., Nature (2000) 408, 541, and references therein 21. Tour, J. M., et al., IEEE Transactions on Nanotechnology (2002) 1, 100 4. Wada, Y., Proc. IEEE (2001) 89, 1147, and references therein 22. Chen, J., et al., Science (1999) 286, 1550 5. Tseng, G. Y., and Ellenbogen, J. C., Science (2001) 294, 1293 23. Seminario, J. M., et al., Search for Minimum Molecular Programmable Units. 6. Huang, Y., et al., Science (2001) 294, 1313 In Proc. IEEE Nanotechnology Conf., (2002) 2, 421 7. Bachtold, A., et al., Science (2001) 294, 1317 24. Seminario, J. M., et al., J. Chem. Phys. (2002) 116, 1671 8. Derycke, V., et al., Nano Lett. (2001) 1, 453 25. Seminario, J. M., et al., Proc. IEEE (2003), in press 9. Liu, X., et al., Appl. Phys. Lett. (2001) 79, 3329 26. Selzer, Y., et al., Presented at United Engineering Foundation Conference on 10. Chen, Y., et al., Nanotechnology (2003) 14, 462 Molecular Electronics, Key West, FL, 2002 11. Chen, Y., et al., Appl. Phys. Lett. (2003) 82, 1610 27. Selzer, Y., et al., Presented at Quilmes Workshop on Molecular Electronics and Quantum Dots, Quilmes, Argentina, 2003 12. Bocian, D. F., et al., DARPA Moletronics Principle Investigator Meeting, McLean VA, July 27-29, 2003 28. Selzer, Y., et al., (2003), unpublished results 13. Tour, J. M., Molecular Electronics: Commercial Insights, Chemistry, Devices, 29. Fang, J., et al., Langmuir (2002) 18, 308 Architecture and Programming, World Scientific, New Jersey (2003) 30. Kim, M. J., et al., Microsc. Microanal. (2002) 8, 1116CD 14. Tour, J. M., Acc. Chem. Res. (2000) 33, 791 31. Gudiksen, M. S., et al., Nature (2002) 415, 617 15. Carroll, R. L., and Gorman, C. B., Angew. Chem. Int. Ed. (2002) 41, 4378 32. Lauhon, L. J., et al., Nature (2002) 420, 57 16. Heath, J. R., et al., Science (1998) 280, 1716 33. Duan, X., et al., Nature (2003) 421, 241 17. Lent, C. S., and Tougaw, P. D., Proc. IEEE (1997) 85, 541 34. Tour, J. M., et al., J. Am. Chem. Soc. (2003) 125 (43), 13279
December 2003 27