Boris Murmann

Bio: Boris Murmann is an academic researcher from Stanford University. The author has contributed to research in topics: CMOS & Amplifier. The author has an hindex of 44, co-authored 231 publications receiving 8486 citations. Previous affiliations of Boris Murmann include University of California, Berkeley & SLAC National Accelerator Laboratory.

Skin electronics from scalable fabrication of an intrinsically stretchable transistor array.

Abstract: Skin-like electronics that can adhere seamlessly to human skin or within the body are highly desirable for applications such as health monitoring, medical treatment, medical implants and biological studies, and for technologies that include human-machine interfaces, soft robotics and augmented reality. Rendering such electronics soft and stretchable-like human skin-would make them more comfortable to wear, and, through increased contact area, would greatly enhance the fidelity of signals acquired from the skin. Structural engineering of rigid inorganic and organic devices has enabled circuit-level stretchability, but this requires sophisticated fabrication techniques and usually suffers from reduced densities of devices within an array. We reasoned that the desired parameters, such as higher mechanical deformability and robustness, improved skin compatibility and higher device density, could be provided by using intrinsically stretchable polymer materials instead. However, the production of intrinsically stretchable materials and devices is still largely in its infancy: such materials have been reported, but functional, intrinsically stretchable electronics have yet to be demonstrated owing to the lack of a scalable fabrication technology. Here we describe a fabrication process that enables high yield and uniformity from a variety of intrinsically stretchable electronic polymers. We demonstrate an intrinsically stretchable polymer transistor array with an unprecedented device density of 347 transistors per square centimetre. The transistors have an average charge-carrier mobility comparable to that of amorphous silicon, varying only slightly (within one order of magnitude) when subjected to 100 per cent strain for 1,000 cycles, without current-voltage hysteresis. Our transistor arrays thus constitute intrinsically stretchable skin electronics, and include an active matrix for sensory arrays, as well as analogue and digital circuit elements. Our process offers a general platform for incorporating other intrinsically stretchable polymer materials, enabling the fabrication of next-generation stretchable skin electronic devices.

Highly stretchable polymer semiconductor films through the nanoconfinement effect

Abstract: Soft and conformable wearable electronics require stretchable semiconductors, but existing ones typically sacrifice charge transport mobility to achieve stretchability. We explore a concept based on the nanoconfinement of polymers to substantially improve the stretchability of polymer semiconductors, without affecting charge transport mobility. The increased polymer chain dynamics under nanoconfinement significantly reduces the modulus of the conjugated polymer and largely delays the onset of crack formation under strain. As a result, our fabricated semiconducting film can be stretched up to 100% strain without affecting mobility, retaining values comparable to that of amorphous silicon. The fully stretchable transistors exhibit high biaxial stretchability with minimal change in on current even when poked with a sharp object. We demonstrate a skinlike finger-wearable driver for a light-emitting diode.

A 12-bit 75-MS/s pipelined ADC using open-loop residue amplification

Abstract: Precision amplifiers dominate the power dissipation in most high-speed pipelined analog-to-digital converters (ADCs). We propose a digital background calibration technique as an enabling element to replace precision amplifiers by simple power-efficient open-loop stages. In the multibit first stage of a 12-bit 75-MSamples/s proof-of-concept prototype, we achieve more than 60% residue amplifier power savings over a conventional implementation. The ADC has been fabricated in a 0.35-/spl mu/m double-poly quadruple-metal CMOS technology and achieves typical differential and integral nonlinearity within 0.5 LSB and 0.9 LSB, respectively. At Nyquist input frequencies, the measured signal-to-noise ratio is 67 dB and the total harmonic distortion is -74 dB. The IC consumes 290 mW at 3 V and occupies 7.9 mm/sup 2/.

Convolutional Neural Networks using Logarithmic Data Representation

Abstract: Recent advances in convolutional neural networks have considered model complexity and hardware efficiency to enable deployment onto embedded systems and mobile devices. For example, it is now well-known that the arithmetic operations of deep networks can be encoded down to 8-bit fixed-point without significant deterioration in performance. However, further reduction in precision down to as low as 3-bit fixed-point results in significant losses in performance. In this paper we propose a new data representation that enables state-of-the-art networks to be encoded to 3 bits with negligible loss in classification performance. To perform this, we take advantage of the fact that the weights and activations in a trained network naturally have non-uniform distributions. Using non-uniform, base-2 logarithmic representation to encode weights, communicate activations, and perform dot-products enables networks to 1) achieve higher classification accuracies than fixed-point at the same resolution and 2) eliminate bulky digital multipliers. Finally, we propose an end-to-end training procedure that uses log representation at 5-bits, which achieves higher final test accuracy than linear at 5-bits.

Matrix-insensitive protein assays push the limits of biosensors in medicine

Abstract: Advances in biosensor technologies for in vitro diagnostics have the potential to transform the practice of medicine. Despite considerable work in the biosensor field, there is still no general sensing platform that can be ubiquitously applied to detect the constellation of biomolecules in diverse clinical samples (for example, serum, urine, cell lysates or saliva) with high sensitivity and large linear dynamic range. A major limitation confounding other technologies is signal distortion that occurs in various matrices due to heterogeneity in ionic strength, pH, temperature and autofluorescence. Here we present a magnetic nanosensor technology that is matrix insensitive yet still capable of rapid, multiplex protein detection with resolution down to attomolar concentrations and extensive linear dynamic range. The matrix insensitivity of our platform to various media demonstrates that our magnetic nanosensor technology can be directly applied to a variety of settings such as molecular biology, clinical diagnostics and biodefense.

Efficient Processing of Deep Neural Networks: A Tutorial and Survey

Abstract: Deep neural networks (DNNs) are currently widely used for many artificial intelligence (AI) applications including computer vision, speech recognition, and robotics. While DNNs deliver state-of-the-art accuracy on many AI tasks, it comes at the cost of high computational complexity. Accordingly, techniques that enable efficient processing of DNNs to improve energy efficiency and throughput without sacrificing application accuracy or increasing hardware cost are critical to the wide deployment of DNNs in AI systems. This article aims to provide a comprehensive tutorial and survey about the recent advances toward the goal of enabling efficient processing of DNNs. Specifically, it will provide an overview of DNNs, discuss various hardware platforms and architectures that support DNNs, and highlight key trends in reducing the computation cost of DNNs either solely via hardware design changes or via joint hardware design and DNN algorithm changes. It will also summarize various development resources that enable researchers and practitioners to quickly get started in this field, and highlight important benchmarking metrics and design considerations that should be used for evaluating the rapidly growing number of DNN hardware designs, optionally including algorithmic codesigns, being proposed in academia and industry. The reader will take away the following concepts from this article: understand the key design considerations for DNNs; be able to evaluate different DNN hardware implementations with benchmarks and comparison metrics; understand the tradeoffs between various hardware architectures and platforms; be able to evaluate the utility of various DNN design techniques for efficient processing; and understand recent implementation trends and opportunities.

Single-Molecule enzyme-linked immunosorbent assay detects serum proteins at subfemtomolar concentrations

Abstract: The ability to detect single protein molecules in blood could accelerate the discovery and use of more sensitive diagnostic biomarkers. To detect low-abundance proteins in blood, we captured them on microscopic beads decorated with specific antibodies and then labeled the immunocomplexes (one or zero labeled target protein molecules per bead) with an enzymatic reporter capable of generating a fluorescent product. After isolating the beads in 50-fl reaction chambers designed to hold only a single bead, we used fluorescence imaging to detect single protein molecules. Our single-molecule enzyme-linked immunosorbent assay (digital ELISA) approach detected as few as approximately 10-20 enzyme-labeled complexes in 100 microl of sample (approximately 10(-19) M) and routinely allowed detection of clinically relevant proteins in serum at concentrations (<10(-15) M) much lower than conventional ELISA. Digital ELISA detected prostate-specific antigen (PSA) in sera from patients who had undergone radical prostatectomy at concentrations as low as 14 fg/ml (0.4 fM).

Skin electronics from scalable fabrication of an intrinsically stretchable transistor array.

Abstract: Skin-like electronics that can adhere seamlessly to human skin or within the body are highly desirable for applications such as health monitoring, medical treatment, medical implants and biological studies, and for technologies that include human-machine interfaces, soft robotics and augmented reality. Rendering such electronics soft and stretchable-like human skin-would make them more comfortable to wear, and, through increased contact area, would greatly enhance the fidelity of signals acquired from the skin. Structural engineering of rigid inorganic and organic devices has enabled circuit-level stretchability, but this requires sophisticated fabrication techniques and usually suffers from reduced densities of devices within an array. We reasoned that the desired parameters, such as higher mechanical deformability and robustness, improved skin compatibility and higher device density, could be provided by using intrinsically stretchable polymer materials instead. However, the production of intrinsically stretchable materials and devices is still largely in its infancy: such materials have been reported, but functional, intrinsically stretchable electronics have yet to be demonstrated owing to the lack of a scalable fabrication technology. Here we describe a fabrication process that enables high yield and uniformity from a variety of intrinsically stretchable electronic polymers. We demonstrate an intrinsically stretchable polymer transistor array with an unprecedented device density of 347 transistors per square centimetre. The transistors have an average charge-carrier mobility comparable to that of amorphous silicon, varying only slightly (within one order of magnitude) when subjected to 100 per cent strain for 1,000 cycles, without current-voltage hysteresis. Our transistor arrays thus constitute intrinsically stretchable skin electronics, and include an active matrix for sensory arrays, as well as analogue and digital circuit elements. Our process offers a general platform for incorporating other intrinsically stretchable polymer materials, enabling the fabrication of next-generation stretchable skin electronic devices.

Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations

Abstract: We introduce a method to train Quantized Neural Networks (QNNs) --- neural networks with extremely low precision (e.g., 1-bit) weights and activations, at run-time. At train-time the quantized weights and activations are used for computing the parameter gradients. During the forward pass, QNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations. As a result, power consumption is expected to be drastically reduced. We trained QNNs over the MNIST, CIFAR-10, SVHN and ImageNet datasets. The resulting QNNs achieve prediction accuracy comparable to their 32-bit counterparts. For example, our quantized version of AlexNet with 1-bit weights and 2-bit activations achieves $51\%$ top-1 accuracy. Moreover, we quantize the parameter gradients to 6-bits as well which enables gradients computation using only bit-wise operation. Quantized recurrent neural networks were tested over the Penn Treebank dataset, and achieved comparable accuracy as their 32-bit counterparts using only 4-bits. Last but not least, we programmed a binary matrix multiplication GPU kernel with which it is possible to run our MNIST QNN 7 times faster than with an unoptimized GPU kernel, without suffering any loss in classification accuracy. The QNN code is available online.