Trading off Energy versus Accuracy in Modern Computing Systems

The slowdown of Moore's law, which has been the driving force of the electronics industry over the last 5 decades, is causing serious problem to Integrated Circuits (ICs) improvements. Technology scaling is becoming more and more complex and fabrication costs are growing exponentially. Furthermore, the energy gains associated to technology scaling are slowing down. Meanwhile, the expected boom of Internet of Things (IoT) devices requires ultra-low power ICs to be able to operate for several years without any user intervention, and energy-efficient computing system on the server side to treat all the gathered data. Approximate computing has emerged as an alternative way to improve energy-efficiency of both, high-performance and low-power computing systems by tolerating small and occasional errors. This energy-accuracy tradeoff can be applied on a wide range of over-engineered applications, particularly those involving human senses such as video and image processing. This thesis first presents an approximate circuit design technique called Gate-Level Pruning, which consists in selectively removing logic gates from any conventional circuit in order to reduce energy consumption, critical path delay, and area occupied on silicon. A Computer Aided Design (CAD) tool has been developed and integrated in the standard digital flow and has been evaluated on several arithmetic circuits, achieving up to 78% energy-delay-area savings. It is then shown how this methodology can be applied on more complex systems made of multiple arithmetic blocks but also memory: the discrete Cosine Transform(DCT), which is a key building block for image and video processing applications. Then, the speculative adder technique is presented. It consists in cutting carry chains to significantly relax the circuit timing constraints', and therefore drastically reduce energy consumption, area and delay. It is shown that this technique leads to errors of different nature than those produced by gate-level pruning. It is therefore worth combining GLP and speculative adders to obtain even higher savings. This has been verified on IEEE-754 floating point units integrated in a 65nm process within a low-power multi-core processor. Silicon measurements show up to 27% power, 36% area and 53% power-area savings. The second part of this thesis introduces software techniques to achieve similar energy-accuracy tradeoffs on commercially available processors. By switching from double precision to single precision floating-point data type and by exploiting vectorization capabilities of modern processors, a factor 2 energy can be saved on a Newton method for solving nonlinear equations. To further investigate the origins of these savings, an energy model based on Energy Per Instructions (EPI) has been built. It turns out that less than 6% of the total energy is consumed by arithmetic operations and that savings are achieved mainly by reducing the amount of data transferred between registers, cache and main memory. One way to reduce those power-hungry data movements is to use application specific hardware accelerators. Unfortunately, a commercial processor cannot embark accelerators for all the possible applications. To that extent, hardware accelerators are implemented on a Field Programmable Gate Array (FPGA) interconnected with a general-purpose processor to further reduce the energy consumption.

Design and Applications of Approximate Circuits by Gate-Level Pruning

Vincent Frédéric Camus, Christian Enz, Jérémy Lucien Maurice Schlachter

Energy-efficiency is a critical concern for many systems, ranging from Internet of things objects and mobile devices to high-performance computers. Moreover, after 40 years of prosperity, Moore's law is starting to show its economic and technical limits. Noticing that many circuits are over-engineered and that many applications are error-resilient or require less precision than offered by the existing hardware, approximate computing has emerged as a potential solution to pursue improvements of digital circuits. In this regard, a technique to systematically tradeoff accuracy in exchange for area, power, and delay savings in digital circuits is proposed: gate-level pruning (GLP). A CAD tool is build and integrated into a standard digital flow to offer a wide range of cost-accuracy tradeoffs for any conventional design. The methodology is first demonstrated on adders, achieving up to 78% energy-delay-area reduction for 10% mean relative error. It is then detailed how this methodology can be applied on a more complex system composed of a multitude of arithmetic blocks and memory: the discrete cosine transform (DCT), which is a key building block for image and video processing applications. Even though arithmetic circuits represent less than 4% of the entire DCT area, it is shown that the GLP technique can lead to 21% energy-delay-area savings over the entire system for a reasonable image quality loss of 24 dB. This significant saving is achieved thanks to the pruned arithmetic circuits, which sets some nodes at constant values, enabling the synthesis tool to further simplify the circuit and memory.

Ieee-Inst Electrical Electronics Engineers Inc2017

Adding limited reconfigurability to superscalar processors

Marc Epalza

For the last thirty years, electronics, at first built with discrete components, and then as Integrated Circuits (IC), have brought diverse and lasting improvements to our quality of life. Examples might include digital calculators, automotive and airplane control assistance, almost all electrical household appliances, and the almost ubiquitous Personal Computer. Application-Specific Integrated Circuits (ASICs) were traditionally used for their high performance and low manufacturing cost, and were designed specifically for a single application with large volumes. But as lower product lifetimes and the pressures of fast marketing increased, ASICs' high design cost pushed for their replacement by Microprocessors. These processors, capable of implementing any functionality through a change in software, are thus often called General Purpose Processors. General purpose processors are used for everyday computing tasks, and found in all personal computers. They are also often used as building blocks for scientific supercomputers. Superscalar processors such as these require ever more processing power to run complex simulations, video games or versatile telecoms services. In the case of embedded applications, e.g. for portable devices, both performance and power consumption must be taken into account. In a bid to adapt a processor to some extent to select applications, fully reconfigurable logic can greatly improve the performance of a processor, since it is shaped for the best possible execution with the available resources. However, as reconfigurable logic is far slower than custom logic, this gain is possible only for some specific applications with large parallelism, after a detailed study of the algorithm. Even though this process can be automated, it still requires large computing resources, and cannot be performed at run time. To reduce the loss in speed compared to custom logic, it is possible to limit the reconfigurability to increase the breadth of applications where performance can be improved. However, as the application space increases, a careful analysis and design of the reconfigurability is required to minimize the speed loss, notably when dynamic reconfiguration is considered. As a case study, we analyze the feasibility of adding limited reconfigurability to the Floating Point Units (FPUs) of a general purpose processor. These rather large units execute all floating point operations, and may also be used for integer multiplication. If an application contains few or infrequent instructions that must be executed by the FPU, this idle hardware only increases power consumption without enhancing performance. This is often the case in non-scientific applications and even many recent and detailed video games which make heavy use of hardware display accelerators for 3D graphics. In a fast multiplier such as can be found in the FPU of a high performance processor, the logic to perform multiplication is a large tree of compressors to add all the partial products together. It is possible to add logic to allow the reconfiguration of part of this tree as several extra Arithmetic and Logic Units (ALU). This requires a detailed timing analysis for both the reconfigurable FPU and the extra ALUs, taking into account effects such as added wires and longer critical paths. Finally, the algorithm to decide when and how to reconfigure must be studied, in terms of eciency and complexity. The results of adding this limited reconfigurability to a mainstream superscalar processor over a large set of compute intensive benchmarks show gains of up to 56% in the best case, with an average gain of 11%. The application to an idealized huge top processor still shows slightly positive average gains, as the limits of available parallelism are reached, bounded by both the application and many of the characteristics of the processor. In all cases, binary compatibility is maintained, allowing the re-use of all existing software. We show that adding limited reconfigurability to a general purpose superscalar processor can produce interesting gains over a wide range of applications while maintaining binary compatibility, and without large modifications to the original design. Limited reconfigurability is worthwhile as it increases the design space, allowing gains to apply to a larger set of applications. These gains are achieved through careful study and optimization of the reconfigurable logic and the decision algorithm.

EPFL2004

Enhancing Design Space Exploration by Extending CPU/GPU Specifications onto FPGAs

Andreas Peter Burg, João Maria Duarte Andrade, Gabriel Falcao Paiva Fernandes, Paolo Ienne, Georgios Karakonstantis, David Novo Bruna, Madhura Purnaprajna

The design cycle for complex special-purpose computing systems is extremely costly and time-consuming. It involves a multiparametric design space exploration for optimization, followed by design verification. Designers of special purpose VLSI implementations often need to explore parameters, such as optimal bitwidth and data representation, through time-consuming Monte Carlo simulations. A prominent example of this simulation-based exploration process is the design of decoders for error correcting systems, such as the Low-Density Parity-Check (LDPC) codes adopted by modern communication standards, which involves thousands of Monte Carlo runs for each design point. Currently, high-performance computing offers a wide set of acceleration options that range from multicore CPUs to Graphics Processing Units (GPUs) and Field Programmable Gate Arrays (FPGAs). The exploitation of diverse target architectures is typically associated with developing multiple code versions, often using distinct programming paradigms. In this context, we evaluate the concept of retargeting a single OpenCL program to multiple platforms, thereby significantly reducing design time. A single OpenCL-based parallel kernel is used without modifications or code tuning on multicore CPUs, GPUs, and FPGAs. We use SOpenCL (Silicon to OpenCL), a tool that automatically converts OpenCL kernels to RTL in order to introduce FPGAs as a potential platform to efficiently execute simulations coded in OpenCL. We use LDPC decoding simulations as a case study. Experimental results were obtained by testing a variety of regular and irregular LDPC codes that range from short/medium (e.g., 8,000 bit) to long length (e.g., 64,800 bit) DVB-S2 codes. We observe that, depending on the design parameters to be simulated, on the dimension and phase of the design, the GPU or FPGA may suit different purposes more conveniently, thus providing different acceleration factors over conventional multicore CPUs.