PHYSICAL REVIEW APPLIED 14, 014097 (2020) Magnetic Field Fingerprinting of Integrated-Circuit Activity with a Quantum Diamond Microscope Matthew J. Turner,1,2 Nicholas Langellier ,1,3 Rachel Bainbridge,4 Dan Walters,4 Srujan Meesala,1,† Thomas M. Babinec ,5 Pauli Kehayias,6 Amir Yacoby,1,5 Evelyn Hu,5 Marko Loncar,ˇ 5 Ronald L. Walsworth ,1,2,3,7,8,9 and Edlyn V. Levine 1,4,* 1 Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA 2 Center for Brain Science, Harvard University, Cambridge, Massachusetts 02138, USA 3 Center for Astrophysics | Harvard & Smithsonian, 60 Garden Street, Cambridge, Massachusetts 02138, USA 4 The MITRE Corporation, Bedford, Massachusetts 01730, USA 5 John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138, USA 6 Sandia National Laboratories, Albuquerque, New Mexico 87123, USA 7 Department of Physics, University of Maryland, College Park, Maryland 20742, USA 8 Department of Electrical and Computer Engineering, University of Maryland, College Park, Maryland 20742, USA 9 Quantum Technology Center, University of Maryland, College Park, Maryland 20742, USA (Received 24 March 2020; revised 24 May 2020; accepted 11 June 2020; published 31 July 2020) Current density distributions in active integrated circuits result in patterns of magnetic fields that contain structural and functional information about the integrated circuit. Magnetic fields pass through standard materials used by the semiconductor industry and provide a powerful means to fingerprint integrated- circuit activity for security and failure analysis applications. Here, we demonstrate high spatial resolution, wide field-of-view, vector magnetic field imaging of static magnetic field emanations from an integrated circuit in different active states using a quantum diamond microscope (QDM). The QDM employs a dense layer of fluorescent nitrogen-vacancy (N-V) quantum defects near the surface of a transparent diamond substrate placed on the integrated circuit to image magnetic fields. We show that QDM imaging achieves a resolution of approximately 10 μm simultaneously for all three vector magnetic field components over the 3.7 × 3.7 mm2 field of view of the diamond. We study activity arising from spatially dependent cur- rent flow in both intact and decapsulated field-programmable gate arrays, and find that QDM images can determine preprogrammed integrated-circuit active states with high fidelity using machine learning classification methods. DOI: 10.1103/PhysRevApplied.14.014097 I. INTRODUCTION and distribution stages [3]. Horizontal integration of the industry has led to contracting of integrated-circuit fabrica- Securing integrated circuits against manufacturing tion, packaging, and testing to offshore facilities, resulting flaws, hardware attacks, and software attacks is of vital in a reduction of secure oversight and quality control importance to the semiconductor industry [1]. Hardware [4]. Additional growth of the secondhand electronics mar- attacks often modify the physical layout of an integrated ket has led to a drastic increase in counterfeit integrated circuit, thereby changing its function. This type of attack circuits [5]. Detection of integrated-circuit tampering or can occur at any stage of the globalized semiconductor counterfeiting has consequently become essential to ensure supply chain, and can range from insertion of malicious hardware can be trusted. Similar issues affect quality con- Trojan circuitry during the design and fabrication stages trol of unintended manufacturing flaws. [2], to modification or counterfeiting during the packaging Magnetic field emanations from integrated circuits afford a powerful means for nondestructive physical test- *[email protected] ing. Magnetic fields are generated by current densities in †Present address: Thomas J. Watson, Sr. Laboratory of integrated circuits resulting from power and clock distri- Applied Physics, California Institute of Technology, Pasadena, bution networks, input and output lines, word and bit lines, California 91125, USA. and switching transistors. These currents are present in 2331-7019/20/14(1)/014097(15) 014097-1 © 2020 American Physical Society MATTHEW J. TURNER et al. PHYS. REV. APPLIED 14, 014097 (2020) all operating logic and memory chips and can be lever- are distinguishable in the QDM images, even with the aged for studying the operational behavior of an integrated diminished magnetic field amplitude and spatial resolution circuit during task execution. In general, the resulting that arise from the large standoff between the diamond and integrated-circuit magnetic fields pass through many stan- the integrated circuit die. We use machine learning meth- dard integrated circuit materials, and will vary spatially ods to demonstrate FPGA operational state classification and temporally in ways that correlate with both integrated- via magnetic field pattern correlation for both decapped circuit architecture and operational state. Thus, combined and intact FPGA QDM images. This result provides an high-resolution and wide-field-of-view mapping of mag- initial demonstration of functional integrated-circuit char- netic fields may yield simultaneous structural and func- acterization via magnetic field fingerprinting. Future work tional information, and may be suitable for identification is required to determine whether and how this approach of malicious circuitry or Trojans [6,7], counterfeit detec- will be useful in areas such as integrated-circuit security tion [8], fault detection [9–11], and manufacturing flaws and failure analysis. [12]. However, leveraging magnetic field emanations is To date, the QDM’s unique combination of magnetic challenging due to the tremendous complexity of circuits field sensitivity, spatial resolution, field of view, and ease integrating billions of transistors of minimum feature sizes of use has allowed it to be used to measure micro- down to tens of nanometers, with interconnects distributed scopic current and magnetization distributions from a across multiple levels of metallization [13]. Multilayered wide variety of sources in both the physical and life sci- metal interconnects and three-dimensional stacking give ences [21–28]. Complementary to scanning techniques rise to complex magnetic field patterns that are difficult for characterizing integrated-circuit magnetic field emana- to invert, and large standoff distances of magnetometers tions, which include wire loops [29], probe antennas [30], reduce amplitudes of magnetic fields and spatial resolu- magnetic force microscopy [11], superconducting quan- tion [14]. tum interference device magnetometers [7], and vapor cell In this paper, we demonstrate how these challenges magnetometers [31], the QDM employs a nonscanning can be approached using a quantum diamond microscope imaging modality [15] that provides simultaneous high- (QDM) [15–17] augmented with machine learning classifi- resolution (micron-scale) and wide-field (millimeter-scale) cation techniques. With the QDM, we perform simultane- vector magnetic imaging, while operating under ambient ous wide field-of-view, high spatial resolution, vector mag- conditions. This capability allows for monitoring of tran- netic field imaging of an operational field-programmable sient behavior over sequential measurements of a magnetic gate array (FPGA). FPGAs are configurable integrated cir- field, providing a means to study correlations in signal pat- cuits that are commonly used for diverse electronics appli- terns that can evolve more quickly than a single-sensor cations. Systematic and controlled variation of the circuit scan time. In addition, the QDM’s simultaneous magnetic activity in the FPGA generates complex magnetic field pat- imaging modality is not subject to the reconstruction errors terns, which we image with the QDM. The QDM employs and drift that can arise from a scanned probe. With these a dense surface layer of fluorescent nitrogen-vacancy distinctive advantages, the QDM technique is a promising (N-V) quantum defects in a macroscopic diamond substrate approach for nondestructive physical testing of integrated placed on the integrated circuit under ambient conditions. circuits. The electronic spins associated with N-V defects have well-established sensitivity to magnetic fields [18–20]. II. EXPERIMENTAL DESIGN We use the QDM to image magnetic fields from both decapsulated (decapped) and through-package (intact) A. QDM experimental setup FPGAs under operational conditions using continuous- A schematic of the QDM is shown in Fig. 1(a).The wave (CW) optically detected magnetic resonance magnetic field sensor consists of a 4 × 4 × 0.5 mm3 dia- (ODMR) N-V spectroscopy. For the decapped FPGA, our mond substrate with a 13 μm surface layer of N-V centers. measurements yield magnetic field maps that are distin- The diamond is placed directly on the integrated circuit guishable between operational states over approximately a with the N-V layer in contact with the integrated-circuit 4 × 4mm2 field of view with a 20 nT noise floor, and a surface. The diamond is grown by Element Six Lim- magnetic field spatial resolution of approximately 10 μm, ited to have an isotopically pure N-V layer consisting of limited by the thickness of the N-V surface layer in the [12C] ∼ 99.999%, [14N] ∼ 27 ppm, and [N-V−] ∼ 2 ppm. diamond and the distance to the nearest metal layer. For Light from a 532 nm, CW laser (Lighthouse Photonics the intact FPGA, the QDM measurements provide mag- Sprout-H-10W) optically addresses the N-V