There is, I think, something ethereal about i —the square root of minus one. I remember first hearing about it at school. It seemed an odd beast at that time—an intruder hovering on the edge of reality.

Usually familiarity dulls this sense of the bizarre, but in the case of i it was the reverse: over the years the sense of its surreal nature intensified. It seemed that it was impossible to write mathematics that described the real world in …

I and i

The organization of behavior: A neuropsychological theory

Many architects believe that major improvements in cost-energy-performance must now come from domain-specific hardware. This paper evaluates a custom ASIC---called a Tensor Processing Unit (TPU)---deployed in datacenters since 2015 that accelerates the inference phase of neural networks (NN). The heart of the TPU is a 65,536 8-bit MAC matrix multiply unit that offers a peak throughput of 92 TeraOps/second (TOPS) and a large (28 MiB) software-managed on-chip memory. The TPU's deterministic execution model is a better match to the 99th-percentile response-time requirement of our NN applications than are the time-varying optimizations of CPUs and GPUs (caches, out-of-order execution, multithreading, multiprocessing, prefetching, ...) that help average throughput more than guaranteed latency. The lack of such features helps explain why, despite having myriad MACs and a big memory, the TPU is relatively small and low power. We compare the TPU to a server-class Intel Haswell CPU and an Nvidia K80 GPU, which are contemporaries deployed in the same datacenters. Our workload, written in the high-level TensorFlow framework, uses production NN applications (MLPs, CNNs, and LSTMs) that represent 95% of our datacenters' NN inference demand. Despite low utilization for some applications, the TPU is on average about 15X - 30X faster than its contemporary GPU or CPU, with TOPS/Watt about 30X - 80X higher. Moreover, using the GPU's GDDR5 memory in the TPU would triple achieved TOPS and raise TOPS/Watt to nearly 70X the GPU and 200X the CPU.

In-Datacenter Performance Analysis of a Tensor Processing Unit

Many architects believe that major improvements in cost-energy-performance must now come from domain-specific hardware. This paper evaluates a custom ASIC---called a Tensor Processing Unit (TPU) --- deployed in datacenters since 2015 that accelerates the inference phase of neural networks (NN). The heart of the TPU is a 65,536 8-bit MAC matrix multiply unit that offers a peak throughput of 92 TeraOps/second (TOPS) and a large (28 MiB) software-managed on-chip memory. The TPU's deterministic execution model is a better match to the 99th-percentile response-time requirement of our NN applications than are the time-varying optimizations of CPUs and GPUs that help average throughput more than guaranteed latency. The lack of such features helps explain why, despite having myriad MACs and a big memory, the TPU is relatively small and low power. We compare the TPU to a server-class Intel Haswell CPU and an Nvidia K80 GPU, which are contemporaries deployed in the same datacenters. Our workload, written in the high-level TensorFlow framework, uses production NN applications (MLPs, CNNs, and LSTMs) that represent 95% of our datacenters' NN inference demand. Despite low utilization for some applications, the TPU is on average about 15X -- 30X faster than its contemporary GPU or CPU, with TOPS/Watt about 30X -- 80X higher. Moreover, using the CPU's GDDR5 memory in the TPU would triple achieved TOPS and raise TOPS/Watt to nearly 70X the GPU and 200X the CPU.

/pdf/in-datacenter-performance-analysis-of-a-tensor-processing-4pmp83zqw4.pdf

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.

Efficient Processing of Deep Neural Networks: A Tutorial and Survey

High leakage current in deep-submicrometer regimes is becoming a significant contributor to power dissipation of CMOS circuits as threshold voltage, channel length, and gate oxide thickness are reduced. Consequently, the identification and modeling of different leakage components is very important for estimation and reduction of leakage power, especially for low-power applications. This paper reviews various transistor intrinsic leakage mechanisms, including weak inversion, drain-induced barrier lowering, gate-induced drain leakage, and gate oxide tunneling. Channel engineering techniques including retrograde well and halo doping are explained as means to manage short-channel effects for continuous scaling of CMOS devices. Finally, the paper explores different circuit techniques to reduce the leakage power consumption.

Leakage current mechanisms and leakage reduction techniques in deep-submicrometer CMOS circuits

In this paper, we have analyzed and modeled failure probabilities (access-time failure, read/write failure, and hold failure) of synchronous random-access memory (SRAM) cells due to process-parameter variations. A method to predict the yield of a memory chip based on the cell-failure probability is proposed. A methodology to statistically design the SRAM cell and the memory organization is proposed using the failure-probability and the yield-prediction models. The developed design strategy statistically sizes different transistors of the SRAM cell and optimizes the number of redundant columns to be used in the SRAM array, to minimize the failure probability of a memory chip under area and leakage constraints. The developed method can be used in an early stage of a design cycle to enhance memory yield in nanometer regime.

Modeling of failure probability and statistical design of SRAM array for yield enhancement in nanoscaled CMOS

This paper presents a programmable and scalable digital neuromorphic architecture based on 3D high-density memory integrated with logic tier for efficient neural computing. The proposed architecture consists of clusters of processing engines, connected by 2D mesh network as a processing tier, which is integrated in 3D with multiple tiers of DRAM. The PE clusters access multiple memory channels (vaults) in parallel. The operating principle, referred to as the memory centric computing, embeds specialized state-machines within the vault controllers of HMC to drive data into the PE clusters. The paper presents the basic architecture of the Neurocube and an analysis of the logic tier synthesized in 28nm and 15nm process technologies. The performance of the Neurocube is evaluated and illustrated through the mapping of a Convolutional Neural Network and estimating the subsequent power and performance for both training and inference.

Neurocube: a programmable digital neuromorphic architecture with high-density 3D memory

Carbon nanotube field-effect transistors (CNFETs) are being extensively studied as possible successors to CMOS. Novel device structures have been fabricated and device simulators have been developed to estimate their performance in a sub-10-nm transistor era. This paper presents a novel method of circuit-compatible modeling of single-walled semiconducting CNFETs in their ultimate performance limit. For the first time, both the I-V and the C-V characteristics of the device have been efficiently modeled for circuit simulations. The model so developed has been used to simulate arithmetic and logic blocks using HSPICE.

A circuit-compatible model of ballistic carbon nanotube field-effect transistors

In this paper, the effect of gate tunneling current in ultra-thin gate oxide MOS devices of effective length (L/sub eff/) of 25nm (oxide thickness=1.1 nm), 50 nm (oxide thickness=1.5 nm) and 90 nm (oxide thickness=2.5 nm) is studied using device simulation. Overall leakage in a stack of transistors is modeled and the opportunities for leakage reduction in the standby mode of operation are explored for scaled technologies. It is shown that, as the contribution of gate leakage relative to the total leakage increases with technology scaling, traditional techniques become ineffective in reducing overall leakage current in a circuit. A novel technique of input vector selection based on the relative contributions of gate and subthreshold leakage to the overall leakage is proposed for reducing total leakage in a circuit. This technique results in 44% savings in total leakage in 50-nm devices compared to the conventional stacking technique.

Saibal Mukhopadhyay

Papers

Leakage current mechanisms and leakage reduction techniques in deep-submicrometer CMOS circuits

Modeling of failure probability and statistical design of SRAM array for yield enhancement in nanoscaled CMOS

Neurocube: a programmable digital neuromorphic architecture with high-density 3D memory

A circuit-compatible model of ballistic carbon nanotube field-effect transistors

Gate leakage reduction for scaled devices using transistor stacking