8 October 2019 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Improving the accuracy of a neural network (NN) usually requires using larger hardware that consumes more energy. However, the error tolerance of NNs and their applications allow approximate computing techniques to be applied to reduce implementation costs. Given that multiplication is the most resource-intensive and power-hungry operation in NNs, more economical approximate multipliers (AMs) can significantly reduce hardware costs. In this article, we show that using AMs can also improve the NN accuracy by introducing noise. We consider two categories of AMs: 1) deliberately designed and 2) Cartesian genetic programing (CGP)-based AMs. The exact multipliers in two representative NNs, a multilayer perceptron (MLP) and a convolutional NN (CNN), are replaced with approximate designs to evaluate their effect on the classification accuracy of the Mixed National Institute of Standards and Technology (MNIST) and Street View House Numbers (SVHN) data sets, respectively. Interestingly, up to 0.63% improvement in the classification accuracy is achieved with reductions of 71.45% and 61.55% in the energy consumption and area, respectively. Finally, the features in an AM are identified that tend to make one design outperform others with respect to NN accuracy. Those features are then used to train a predictor that indicates how well an AM is likely to work in an NN.
20 October 2020 in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
Libraries of approximate circuits are composed of fully characterized digital circuits that can be used as building blocks of energy-efficient implementations of hardware accelerators. They can be employed not only to speed up the accelerator development but also to analyze how an accelerator responds to introducing various approximate operations. In this paper, we present a methodology that automatically builds comprehensive libraries of approximate circuits with desired properties. Target approximate circuits are generated using Cartesian genetic programming. In addition to extending the EvoApprox8b library that contains common approximate arithmetic circuits, we show how to generate more specific approximate circuits; in particular, MxN-bit approximate multipliers that exhibit promising results when deployed in convolutional neural networks. By means of the evolved approximate multipliers, we perform a detailed error resilience analysis of five different ResNet networks. We identify the convolutional layers that are good candidates for adopting the approximate multipliers and suggest particular approximate multipliers whose application can lead to the best trade-offs between the classification accuracy and energy requirements. Experiments are reported for CIFAR-10 and CIFAR-100 data sets.
30 July 2018 in IEEE Transactions on Very Large Scale Integration (VLSI) Systems
Approximate computing exploits the fact that many applications are inherently error resilient. In order to reduce power consumption, approximate circuits such as multipliers have been employed in these applications. However, most current approximate multipliers are based on ad hoc circuit structures and, for automated circuit approximation methods, large efficient designs are difficult to find due to the increased search space. Moreover, existing design methods do not typically provide sufficient formal guarantees in terms of error if large approximate multipliers are constructed. To address these challenges, this brief introduces a general and efficient method for constructing large high-quality approximate multipliers with respect to the objectives formulated in terms of the power-delay product and a provable error bound. This is demonstrated by means of a comparative evaluation of approximate 16-bit multipliers constructed by the proposed method and other methods in the literature.
10 December 2019 in IEEE Access
Approximate circuits are digital circuits that are intentionally designed in such a way that the specification is violated in terms of functionality in order to obtain some improvements in power consumption, performance or area, in comparison with fully functional circuits. To design the approximate circuits, the synthesis tools rely on the availability of a procedure checking, whether the synthesized circuits meet a specification and/or provides information about circuit quality. Compared to the traditional circuit design flow, the nature of the approximate circuits involves replacing the strict functional equivalence checking with a more advanced approach that enables us to quantify or guarantee the degree of similarity. The most common technique is to employ a circuit simulator for analysing responses for all input vectors. This approach allows us to simultaneously perform checking and quality assessment, but the exhaustive enumeration of the input vectors is tractable only for a small number of inputs. To avoid excessive run-times, a subset of all possible input vectors is typically used for complex circuits. This causes us, however, to lose the ability to guarantee that the quality of the synthesized circuits is within an acceptable range given in the specification. The main goal of this paper is to show how to adopt formal methods such as binary decision diagrams and satisfiability solvers for exhaustive analysis of approximate circuits without explicit enumeration of all input vectors. We survey the methods for exact computation of the most important error parameters used in the context of approximate computing, propose improved algorithms and provide a detailed analysis of their performance. The methods are benchmarked on a large set of key approximate circuits consisting of nearly 2,000 unique arithmetic instances with 8-, 12-, 16-, and 32-bit operands which helps us to identify the best algorithm and method for computation of a desired error parameter.
13 March 2011 in Genetic Programming and Evolvable Machines