Search

Semiconductor innovation drives improvements to technologies that are critical to modern society. The country that successfully accelerates semiconductor innovation is positioned to lead future semiconductor-driven industries and benefit from the resulting economic growth. It is our view that a next generation infrastructure is necessary to accelerate and enhance semiconductor innovation in the U.S. In this paper, we propose such an advanced infrastructure composed of a national network of facilities with enhancements in technology and business models. These enhancements enable application-driven and challenge-based research and development, and ensure that facilities are accessible and sustainable. The main tenets are: a challenge-driven operational model, a next-generation infrastructure to serve that operational model, technology innovations needed for advanced facilities to speed up learning cycles, and innovative cost-effective business models for sustainability. Ultimately, the expected outcomes of such a participatory, scalable, and sustainable nation-level advanced infrastructure will have tremendous impact on government, industry, and academia alike.

This paper studies the merits of applying log-gradient input images to convolutional neural networks (CNNs) for tinyML computer vision (CV). We show that log gradients enable: (i) aggressive 1.5-bit quantization of first-layer inputs, (ii) potential CNN resource reductions, and (iii) inherent robustness to illumination changes (1.7% accuracy loss across 1/32...8 brightness variation vs. up to 10% for JPEG). We establish these results using the PASCAL RAW image data set and through a combination of experiments using neural architecture search and a fixed three-layer network. The latter reveal that training on log-gradient images leads to higher filter similarity, making the CNN more prunable. The combined benefits of aggressive first-layer quantization, CNN resource reductions, and operation without tight exposure control and image signal processing (ISP) are helpful for pushing tinyML CV toward its ultimate efficiency limits., Comment: 8 pages

The recent success of neural networks for solving difficult decision tasks has incentivized incorporating smart decision making "at the edge." However, this work has traditionally focused on neural network inference, rather than training, due to memory and compute limitations, especially in emerging non-volatile memory systems, where writes are energetically costly and reduce lifespan. Yet, the ability to train at the edge is becoming increasingly important as it enables real-time adaptability to device drift and environmental variation, user customization, and federated learning across devices. In this work, we address two key challenges for training on edge devices with non-volatile memory: low write density and low auxiliary memory. We present a low-rank training scheme that addresses these challenges while maintaining computational efficiency. We then demonstrate the technique on a representative convolutional neural network across several adaptation problems, where it out-performs standard SGD both in accuracy and in number of weight writes.

Electronic biosensors are a natural fit for field-deployable diagnostic devices, because they can be miniaturized, mass produced, and integrated with circuitry. Unfortunately, progress in the development of such platforms has been hindered by the fact that mobile ions present in biological samples screen charges from the target molecule, greatly reducing sensor sensitivity. Under physiological conditions, the thickness of the resulting electric double layer is less than 1 nm, and it has generally been assumed that electronic detection beyond this distance is virtually impossible. However, a few recently-described sensor design strategies seem to defy this conventional wisdom, exploiting the physics of electrical double layers in ways that traditional models do not capture. In the first strategy, charge screening is decreased by constraining the space in which double layers can form. The second strategy uses external stimuli to prevent double layers from reaching equilibrium, thereby effectively reducing charge screening. The goal of this article is to describe these relatively new concepts, and to offer theoretical insights into mechanisms that may enable electronic biosensing beyond the double-layer. If these concepts can be further developed and translated into practical electronic biosensors, we foresee exciting opportunities for the next generation of diagnostic technologies.

Batch Normalization (BatchNorm) is commonly used in Convolutional Neural Networks (CNNs) to improve training speed and stability. However, there is still limited consensus on why this technique is effective. This paper uses concepts from the traditional adaptive filter domain to provide insight into the dynamics and inner workings of BatchNorm. First, we show that the convolution weight updates have natural modes whose stability and convergence speed are tied to the eigenvalues of the input autocorrelation matrices, which are controlled by BatchNorm through the convolution layers' channel-wise structure. Furthermore, our experiments demonstrate that the speed and stability benefits are distinct effects. At low learning rates, it is BatchNorm's amplification of the smallest eigenvalues that improves convergence speed, while at high learning rates, it is BatchNorm's suppression of the largest eigenvalues that ensures stability. Lastly, we prove that in the first training step, when normalization is needed most, BatchNorm satisfies the same optimization as Normalized Least Mean Square (NLMS), while it continues to approximate this condition in subsequent steps. The analyses provided in this paper lay the groundwork for gaining further insight into the operation of modern neural network structures using adaptive filter theory., Comment: Presented at Asilomar Conference on Signals, Systems, and Computers, 2020

This paper introduces BinarEye: a digital processor for always-on Binary Convolutional Neural Networks. The chip maximizes data reuse through a Neuron Array exploiting local weight Flip-Flops. It stores full network models and feature maps and hence requires no off-chip bandwidth, which leads to a 230 1b-TOPS/W peak efficiency. Its 3 levels of flexibility - (a) weight reconfiguration, (b) a programmable network depth and (c) a programmable network width - allow trading energy for accuracy depending on the task's requirements. BinarEye's full system input-to-label energy consumption ranges from 14.4uJ/f for 86% CIFAR-10 and 98% owner recognition down to 0.92uJ/f for 94% face detection at up to 1700 frames per second. This is 3-12-70x more efficient than the state-of-the-art at on-par accuracy., Comment: Presented at the 2018 IEEE Custom Integrated Circuits Conference (CICC). Presentation is available here: https://www.researchgate.net/publication/324452819_Presentation_on_Binareye_at_CICC

We introduce a classical potentiostatic feedback mechanism that attenuates the dissipation in a quantum system arising from coupling to the surrounding thermodynamic bath, preserving the inter-state interference in an electronic excitation transfer (EET) process. A three-terminal potentiostat device applies a low-noise voltage bias to the terminals of the EET system and reduces the physical coupling between the quantum system and its environment. We introduce a classical equivalent circuit to model the environment-coupled excitation transfer in an elementary two-state system. This model provides qualitative insight into how classical feedback action affects the transition probabilities between the states and selectively reduces the dissipative coupling for one of the vibronic energy levels of the transfer system. Furthermore, we show that negative feedback results in persistent spectral coherence between the energy level of the decoupled state and the vibronic levels of the complementary state, making the decoupled vibronic channel a probe for characterizing the vibronic structure of the complementary channel of the EET system., Comment: 4 figures

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., Comment: 10 pages, 7 figures