Approximate computing has been considered to improve the accuracy-performance trade-off in error-tolerant applications. For many of these applications, multiplication is a key arithmetic operation. Given that approximate compressors are a key element in the design of power-efficient approximate multipliers, we first propose an initial approximate 4:2 compressor that introduces a rather large error to the output. However, the number of faulty rows in the compressor’s truth table is significantly reduced by encoding its inputs using generate and propagate signals. Based on this improved compressor, two 4×4 multipliers are designed with different accuracies and then are used as building blocks for scaling up to 16×16 and 32×32 multipliers. According to the mean relative error distance (MRED), the most accurate of the proposed 16×16 unsigned designs has a 44% smaller power-delay product (PDP) compared to other designs with comparable accuracy. The radix-4 signed Booth multiplier constructed using the proposed compressor achieves a 52% reduction in the PDP-MRED product compared to other approximate Booth multipliers with comparable accuracy. The proposed multipliers outperform other approximate designs in image sharpening and joint photographic experts group (JPEG) applications by achieving higher quality outputs with lower power consumptions. For the first time, we show the applicability and practicality of approximate multipliers in multiple-input multiple-output (MIMO) antenna communication systems with error control coding.
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.
Multiplication is the most resource-hungry operation in neural networks (NNs). Logarithmic multipliers (LMs) simplify multiplication to shift and addition operations and thus reduce the energy consumption. Since implementing the logarithm in a compact circuit often introduces approximation, some accuracy loss is inevitable in LMs. However, this loss is tolerated by the inherent error tolerance of NNs and their associated applications. This article proposes an improved logarithmic multiplier (ILM) that, unlike existing designs, rounds both inputs to their nearest powers of two by using a proposed nearest-one detector (NOD) circuit. Considering that the output of the NOD uses a one-hot representation, some entries in the truth table of a conventional adder cannot occur. Hence, a compact adder is designed for the reduced truth table. The 8×8 ILM achieves up to 17.48% saving in power consumption compared to a recent LM in the literature while being almost 15% more accurate. Moreover, the evaluation of the ILM for two benchmark NN workloads shows up to 21.85% reduction in energy consumption compared to the NNs implemented with other LMs. Interestingly, using the ILM increases the classification accuracy of the considered NNs by up to 1.4% compared to a NN implementation that uses exact multipliers.
24 April 2020 in IEEE Transactions on Emerging Topics in Computing
The squaring function is widely used in Digital Signal Processing (DSP). There are many DSP applications with noisy inputs for which simplifying approximations of the squaring function implementation have a minor impact on the output quality, while permitting significant reductions in the hardware cost. This article proposes a Low-Error Squaring Function (LESF) and its low-power hardware implementation. Unlike the existing logarithmic squaring functions, LESF benefits from a double-sided error distribution and, consequently, error cancellation in larger calculations. LESF approximates a base-2 logarithmic function with a piecewise linear polynomial, i.e. $\mathrm{log_2}f(x) \approx ax+b$ . Since input $b$ in this sum is a constant, LESF replaces the conventional full-adder with a compact specialized adder for hardware efficiency. Our simulation results show that the 16-bit LESF is 23.23% more accurate (assuming the mean relative error distance metric) than the baseline Mitchell approximate logarithmic squaring function while being 1.8× faster and 39% more energy-efficient. LESF and other logarithmic squaring functions are evaluated for the square-law detector application. LESF is shown to be more than 3× more accurate in this application (with respect to the Euclidean distance metric) than the next most accurate design in the literature, which uses an iterative error compensation technique.
Considering that practical wireless sensor networks are mostly built upon energy‐constrained sensor nodes, network lifetime becomes a key deployment factor for sustainability of such networks initiating an attempt for prolonging the lifetime. Adopting a new multi‐mode switching protocol for sensor nodes between active and sleep states, in this study, the authors propose a new mathematical method using an embedded simulation‐based computing of an energy‐efficient strategy performance metric to maintain resiliency of the network against side effects appearing in duty‐cycling protocols. Here, the authors use a Monte Carlo simulation to estimate network reliability. Then the proposed mathematical model enables us to build a decision for optimised performance upon trade‐offs between consumed energy, reliability, coverage intensity and end‐to‐end delays by locating a network's operational points. The authors show the trade‐off operational significance of the model against the percentage of sleeping probability via some design factors including figure of merit.
22 March 2021 in IET Computers & Digital Techniques
The logarithmic number system (LNS) can be used to simplify the computation of arithmetic functions, such as multiplication. This article proposes three leading‐one detectors (LODs) to speed up the binary logarithm calculation in the LNS. The first LOD (LOD I) uses a single fixed value to approximate the d least significant bits (LSBs) in the outputs of the LOD. The second design (LOD II) partitions the d LSBs into smaller fields and uses a multiplexer to select the closest approximation to the exact value. These two LODs help with error cancellation as they introduce signed errors for inputs N < 2 d . Additionally, a scaling scheme is proposed that scales up the input N < 2 d to avoid large approximation errors. Finally, an improved exact LOD (LOD III) is proposed that only passes half of the input N to the LOD; the more significant half is passed if there is at least one ‘1’ in that half; otherwise, the less significant half is passed. Our simulation results show that the 32‐bit LOD III can be up to 2.8× more energy‐efficient than existing designs in the literature. The Mitchell logarithmic multiplier and a neural network are considered to further illustrate the practicality of the proposed designs.