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

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

There are two general approaches to developing artificial general intelligence (AGI)1: computer-science-oriented and neuroscience-oriented. Because of the fundamental differences in their formulations and coding schemes, these two approaches rely on distinct and incompatible platforms2–8, retarding the development of AGI. A general platform that could support the prevailing computer-science-based artificial neural networks as well as neuroscience-inspired models and algorithms is highly desirable. Here we present the Tianjic chip, which integrates the two approaches to provide a hybrid, synergistic platform. The Tianjic chip adopts a many-core architecture, reconfigurable building blocks and a streamlined dataflow with hybrid coding schemes, and can not only accommodate computer-science-based machine-learning algorithms, but also easily implement brain-inspired circuits and several coding schemes. Using just one chip, we demonstrate the simultaneous processing of versatile algorithms and models in an unmanned bicycle system, realizing real-time object detection, tracking, voice control, obstacle avoidance and balance control. Our study is expected to stimulate AGI development by paving the way to more generalized hardware platforms. The ‘Tianjic’ hybrid electronic chip combines neuroscience-oriented and computer-science-oriented approaches to artificial general intelligence, demonstrated by controlling an unmanned bicycle.

Towards artificial general intelligence with hybrid Tianjic chip architecture.

Hyperscale datacenter providers have struggled to balance the growing need for specialized hardware (efficiency) with the economic benefits of homogeneity (manageability) In this paper we propose a new cloud architecture that uses reconfigurable logic to accelerate both network plane functions and applications This Configurable Cloud architecture places a layer of reconfigurable logic (FPGAs) between the network switches and the servers, enabling network flows to be programmably transformed at line rate, enabling acceleration of local applications running on the server, and enabling the FPGAs to communicate directly, at datacenter scale, to harvest remote FPGAs unused by their local servers We deployed this design over a production server bed, and show how it can be used for both service acceleration (Web search ranking) and network acceleration (encryption of data in transit at high-speeds) This architecture is much more scalable than prior work which used secondary rack-scale networks for inter-FPGA communication By coupling to the network plane, direct FPGA-to-FPGA messages can be achieved at comparable latency to previous work, without the secondary network Additionally, the scale of direct inter-FPGA messaging is much larger The average round-trip latencies observed in our measurements among 24, 1000, and 250,000 machines are under 3, 9, and 20 microseconds, respectively The Configurable Cloud architecture has been deployed at hyperscale in Microsoft's production datacenters worldwide

/pdf/a-cloud-scale-acceleration-architecture-8ejpq6fh99.pdf

A cloud-scale acceleration architecture

To advance datacenter capabilities beyond what commodity server designs can provide, the authors designed and built a composable, reconfigurable fabric to accelerate large-scale software services. Each instantiation of the fabric consists of a 6 x 8 2D torus of high-end field-programmable gate arrays (FPGAs) embedded into a half-rack of 48 servers. The authors deployed the reconfigurable fabric in a bed of 1,632 servers and FPGAs in a production datacenter and successfully used it to accelerate the ranking portion of the Bing Web search engine by nearly a factor of two.

/pdf/a-reconfigurable-fabric-for-accelerating-large-scale-4cqgkcjehf.pdf

A Reconfigurable Fabric for Accelerating Large-Scale Datacenter Services

A two-way set associative cache memory system for a parallel-pipelined computer system uses separate queue structures to hold main memory fetch and store requests generated by the central processing unit (CPU) A memory access unit, coupled between the cache memory system and the CPU selects the next request to be processed by the main memory from between the requests at the heads of the fetch and store queues The request at the head of the fetch queue is preferred over the request at the head of the store queue unless the memory partition to be used by the fetch request is still busy with a previous request while the partition to be used by the store request is idle Data retrieved from the main memory replaces data in the cache according to an algorithm that prefers empty pages within a set to pages that contain data and prefers pages that do not have pending update requests scheduled to pages that do have pending update requests scheduled In the event that only pages having pending update requests are found, input requests to the cache are inhibited until at least one fetch request for a page in the set is completed and the page is no longer marked as having a pending update request

Method and apparatus for cache memory access with separate fetch and store queues

Hyperscale datacenter providers have struggled to balance the growing need for specialized hardware with the economic benefits of homogeneity. The Configurable Cloud datacenter architecture introduces a layer of reconfigurable logic (FPGAs) between the network switches and servers. This enables line-rate transformation of network packets, acceleration of local applications running on the server, and direct communication among FPGAs, at datacenter scale. This low latency, ubiquitous communication enables deployment of hardware services spanning any number of FPGAs to be used and shared quickly and efficiently by services of any scale throughout the datacenter. The authors deploy this design over a production server bed and show how it can be used to accelerate applications that were explicitly ported to FPGAs and support hardware-first services. It can even accelerate applications without any application-specific FPGA code being written. The Configurable Cloud architecture has been deployed at hyperscale in Microsoft's production datacenters worldwide.

Configurable Clouds

A display device includes a waveguide, a reconfigurable phase mask, and a controller. The controller dynamically reconfigures the reconfigurable phase mask so as to modulate display light in accordance with a detected position of an eye and/or a parameter for a shape of the waveguide. The waveguide transmits the modulated display light.

Image correction using reconfigurable phase mask

The subject disclosure is directed towards emulating a data network interface for operating an interconnect controller in accordance with a networking standard The emulated data network interface provides address resolution data associated with a destination, within a converged interconnect fabric, for application data The address resolution data identifies a communication path that is compatible with the interconnect controller The application data conforms to the networking standard The emulated data network interface instructs the interconnect controller to route the application data to the destination via the compatible communication path

Stephen F. Heil

Papers

A Reconfigurable Fabric for Accelerating Large-Scale Datacenter Services

Method and apparatus for cache memory access with separate fetch and store queues

Configurable Clouds

Image correction using reconfigurable phase mask

Providing Network Capability Over A Converged Interconnect Fabric