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

A number of recent efforts have attempted to design accelerators for popular machine learning algorithms, such as those involving convolutional and deep neural networks (CNNs and DNNs). These algorithms typically involve a large number of multiply-accumulate (dot-product) operations. A recent project, DaDianNao, adopts a near data processing approach, where a specialized neural functional unit performs all the digital arithmetic operations and receives input weights from adjacent eDRAM banks.This work explores an in-situ processing approach, where memristor crossbar arrays not only store input weights, but are also used to perform dot-product operations in an analog manner. While the use of crossbar memory as an analog dot-product engine is well known, no prior work has designed or characterized a full-fledged accelerator based on crossbars. In particular, our work makes the following contributions: (i) We design a pipelined architecture, with some crossbars dedicated for each neural network layer, and eDRAM buffers that aggregate data between pipeline stages. (ii) We define new data encoding techniques that are amenable to analog computations and that can reduce the high overheads of analog-to-digital conversion (ADC). (iii) We define the many supporting digital components required in an analog CNN accelerator and carry out a design space exploration to identify the best balance of memristor storage/compute, ADCs, and eDRAM storage on a chip. On a suite of CNN and DNN workloads, the proposed ISAAC architecture yields improvements of 14.8×, 5.5×, and 7.5× in throughput, energy, and computational density (respectively), relative to the state-of-the-art DaDianNao architecture.

ISAAC: a convolutional neural network accelerator with in-situ analog arithmetic in crossbars

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 towards 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 co-designs, 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 trade-offs 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.

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

This paper proposes a power-aware digital feedforward neural network platform that utilizes the backpropagation algorithm during training to enable energy-quality trade-off. Given a quality constraint, the proposed approach identifies a set of synaptic weights for approximation in a neural network. The approach selects synapses with small impact on output error, estimated by the backpropagation algorithm, for approximation. The approximations are achieved by a coupled software (reduced bit-width) and hardware (approximate multiplication in the processing engine) based design approaches. The full-chip design in 130nm CMOS shows, compared to a baseline accurate design, the proposed approach reduces system power by ∼38% with 0.4% lower recognition accuracy in a classification problem.

A power-aware digital feedforward neural network platform with backpropagation driven approximate synapses

This paper proposes that approximation by reducing bit-precision and using inexact multiplier can save power consumption of digital multilayer perceptron accelerator during the classification of MNIST (inference) with negligible accuracy degradation. Based on the error sensitivity precomputed during the training, synaptic weights with less sensitivity are approximated. Under given bit-precision modes, our proposed algorithm determines bit precision for all synapse to minimize power consumption for given target accuracy. For entire network, earlier layer can be more approximated since it has lower error sensitivity. Proposed algorithm can save power 57.4 percent while accuracy is degraded about 1.7 percent. After approximation, retraining with few iterations can improve the accuracy while maintaining power consumption. The impact of different training conditions on the approximation is also studied. Training with small quantization error (less bit precision) allows more power saving in inference. It also shows that enough number of iteration during the training is important for approximation in inference. Network with more layers is more sensitive to the approximation.

A Power-Aware Digital Multilayer Perceptron Accelerator with On-Chip Training Based on Approximate Computing

Neural networks generally require significant memory capacity/bandwidth to store/access a large number of synaptic weights. This paper presents an application of JPEG image encoding to compress the weights by exploiting the spatial locality and smoothness of the weight matrix. To minimize the loss of accuracy due to JPEG encoding, we propose to adaptively control the quantization factor of the JPEG algorithm depending on the error-sensitivity (gradient) of each weight. With the adaptive compression technique, the weight blocks with higher sensitivity are compressed less for higher accuracy. The adaptive compression reduces memory requirement, which in turn results in higher performance and lower energy of neural network hardware. The simulation for inference hardware for multilayer perceptron with the MNIST dataset shows up to 42X compression with less than 1% loss of recognition accuracy, resulting in 3X higher effective memory bandwidth and ∼19X lower system energy.

Adaptive weight compression for memory-efficient neural networks

This paper investigates the power and performance trade-offs associated with integrating deep neural network (DNN) computation in an image sensor. The paper presents the design of Neurosensor–a CMOS image sensor with 3-D stacking of pixel array, read-out circuits, memory, and computing logic for DNN. The analysis shows integrating DNN reduces transmit latency (and energy), but at the expense of processing and memory access latency (and energy). Hence, given a specific DNN and transmission bandwidth, there exist an optimal number of layers that should be computed in the sensor to maximize energy-efficiency. In general, it is often more efficient to integrate memory within the sensor stack and/or implement only the feature extraction layers on the sensor, and optimized configurations can achieve up to $90\times $ improvement in energy efficiency compared to the baseline. Further, coupled power, thermal, and noise simulation demonstrates that integrating DNN computation can increase pixel-array temperature resulting in higher noise, and hence, lower classification accuracy.

Duckhwan Kim

Papers

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

A power-aware digital feedforward neural network platform with backpropagation driven approximate synapses

A Power-Aware Digital Multilayer Perceptron Accelerator with On-Chip Training Based on Approximate Computing

Adaptive weight compression for memory-efficient neural networks

3-D Stacked Image Sensor With Deep Neural Network Computation