A Study on Efficient Designs of Approximate Arithmetic Circuits
dc.contributor.advisor | Ko, Seok-Bum | |
dc.contributor.committeeMember | Dinh, Anh | |
dc.contributor.committeeMember | Chen, Li | |
dc.contributor.committeeMember | Karki, Rajesh | |
dc.contributor.committeeMember | Wu, Fangxiang | |
dc.creator | Venkatachalam, Suganthi | |
dc.date.accessioned | 2019-01-03T21:55:15Z | |
dc.date.available | 2021-01-03T06:05:07Z | |
dc.date.created | 2018-12 | |
dc.date.issued | 2019-01-03 | |
dc.date.submitted | December 2018 | |
dc.date.updated | 2019-01-03T21:55:16Z | |
dc.description.abstract | Approximate computing is a popular field where accuracy is traded with energy. It can benefit applications such as multimedia, mobile computing and machine learning which are inherently error resilient. Error introduced in these applications to a certain degree is beyond human perception. This flexibility can be exploited to design area, delay and power efficient architectures. However, care must be taken on how approximation compromises the correctness of results. This research work aims to provide approximate hardware architectures with error metrics and design metrics analyzed and their effects in image processing applications. Firstly, we study and propose unsigned array multipliers based on probability statistics and with approximate 4-2 compressors, full adders and half adders. This work deals with a new design approach for approximation of multipliers. The partial products of the multiplier are altered to introduce varying probability terms. Logic complexity of approximation is varied for the accumulation of altered partial products based on their probability. The proposed approximation is utilized in two variants of 16-bit multipliers. Synthesis results reveal that two proposed multipliers achieve power savings of 72% and 38% respectively compared to an exact multiplier. They have better precision when compared to existing approximate multipliers. Mean relative error distance (MRED) figures are as low as 7.6% and 0.02% for the proposed approximate multipliers, which are better than the previous state-of-the-art works. Performance of the proposed multipliers is evaluated with geometric mean filtering application, where one of the proposed models achieves the highest peak signal to noise ratio (PSNR). Second, approximation is proposed for signed Booth multiplication. Approximation is introduced in partial product generation and partial product accumulation circuits. In this work, three multipliers (ABM-M1, ABM-M2, and ABM-M3) are proposed in which the modified Booth algorithm is approximated. In all three designs, approximate Booth partial product generators are designed with different variations of approximation. The approximations are performed by reducing the logic complexity of the Booth partial product generator, and the accumulation of partial products is slightly modified to improve circuit performance. Compared to the exact Booth multiplier, ABM-M1 achieves up to 15% reduction in power consumption with an MRED value of 7.9 × 10⁻⁴. ABM-M2 has power savings of up to 60% with an MRED of 1.1 × 10⁻¹. ABM-M3 has power savings of up to 50% with an MRED of 3.4 × 10⁻³. Compared to existing approximate Booth multipliers, the proposed multipliers ABM-M1 and ABM-M3 achieve up to a 41% reduction in power consumption while exhibiting very similar error metrics. Image multiplication and matrix multiplication are used as case studies to illustrate the high performance of the proposed approximate multipliers. Third, distributed arithmetic based sum of products units approximation is analyzed. Sum of products units are key elements in many digital signal processing applications. Three approximate sum of products models which are based on distributed arithmetic are proposed. They are designed for different levels of accuracy. First model of approximate sum of products achieves an improvement up to 64% on area and 70% on power, when compared to conventional unit. Other two models provide an improvement of 32% and 48% on area and 54% and 58% on power, respectively, with a reduced error rate compared to the first model. Third model achieves MRED and normalized mean error distance (NMED) as low as 0.05% and 0.009%. Performance of approximate units is evaluated with a noisy image smoothing application, where the proposed models are capable of achieving higher PSNR than existing state of the art techniques. Fourth, approximation is applied in division architecture. Two approximation models are proposed for restoring divider. In the first design, approximation is performed at circuit level, where approximate divider cells are utilized in place of exact ones by simplifying the logic equations. In the second model, restoring divider is analyzed strategically and number of restoring divider cells are reduced by finding the portions of divisor and dividend with significant information. An approximation factor 'p' is used in both designs. In model 1, the design with p=8 has a 58% reduction in both area and power consumption compared to exact design, with a Q-MRED of 1.909 × 10⁻² and Q-NMED of 0.449 × 10⁻². The second model with an approximation factor p=4 has 54% area savings and 62% power savings compared to exact design. The proposed models are found to have better error metrics compared to existing designs, with better performance at similar error values. A change detection image processing application is used for real time assessment of proposed and existing approximate dividers and one of the models achieves a PSNR of 54.27 dB. | |
dc.format.mimetype | application/pdf | |
dc.identifier.uri | http://hdl.handle.net/10388/11699 | |
dc.subject | approximate computing | |
dc.subject | arithmetic circuits | |
dc.title | A Study on Efficient Designs of Approximate Arithmetic Circuits | |
dc.type | Thesis | |
dc.type.material | text | |
local.embargo.terms | 2021-01-03 | |
thesis.degree.department | Electrical and Computer Engineering | |
thesis.degree.discipline | Electrical Engineering | |
thesis.degree.grantor | University of Saskatchewan | |
thesis.degree.level | Doctoral | |
thesis.degree.name | Doctor of Philosophy (Ph.D.) |