SIMDive: Approximate SIMD Soft Multiplier-Divider for FPGAs with Tunable Accuracy

Abstract

The ever-increasing quest for data-level parallelism and variable precision in ubiquitous multimedia and Deep Neural Network (DNN) applications has motivated the use of Single Instruction, Multiple Data (SIMD) architectures. To alleviate energy as their main resource constraint, approximate computing has re-emerged, albeit mainly specialized for their Application-Specific Integrated Circuit (ASIC) implementations. This paper, presents for the first time, an SIMD architecture based on novel multiplier and divider with tunable accuracy, targeted for Field-Programmable Gate Arrays (FPGAs). The proposed hybrid architecture implements Mitchell's algorithms and supports precision variability from 8 to 32 bits. Experimental results obtained from Vivado, multimedia and DNN applications indicate superiority of proposed architecture (both SISD and SIMD) over accurate and state-of-the-art approximate counterparts. In particular, the proposed SISD divider outperforms the accurate Intellectual Property (IP) divider provided by Xilinx with 4× higher speed and 4.6×less energy and tolerating only <0.8% error. Moreover, the proposed SIMD multiplier-divider supersede accurate SIMD multiplier by achieving up to 26%, 45%, 36%, and 56% improvement in area, throughput, power, and energy, respectively.

BioCare: An Energy-Efficient CGRA for Bio-Signal Processing at the Edge

“…This highlights the demand for a generic approximate SIMD ALU, that can also be utilized in health-monitoring platforms. Moreover, while Add/Mul are frequent functions in bio-signal processing workloads, [9], [15], [18] and our evaluations show that long latency/high power of division, not only limit application speed, but also consumes considerable portion of ALU area/energy (Fig. 1b).…”

“…However, such CGRAs suffer from high area, stems from multiple datapath in PEs. Amortizing this penalty, few works have narrowed their focus to design approximate SIMD Mul and/or Div [15], [19].…”

“…We adopt Mul/Div approximation of SIMDive [15], the rationale behind which is multi-fold: 1 it has proved having a superior resource-accuracy trade-off over SoAs [20], plus its error-strategy is adjustable and independent from Mul/Div size. 2 Its errors are unbiased: they are centered nearly symmetrical around zero, therefore, can cancel out each others in consecutive bio-signal kernels (having mostly Add/Mul) and prevent PSNR degradation.…”

“…The concept of SIMD is recently exploited by e.g., Xilinx [16] and Intel [17], to provide support for precision-variability in a single unit. Such efforts, however, are restricted to two Muls with a common operand, or Mul/Div [15], in FPGAs. This highlights the demand for a generic approximate SIMD ALU, that can also be utilized in health-monitoring platforms.…”

“…Such a light-weight SIMD architecture can perfectly fit for Mul-exhaustive bio-signal processing, while also allows better utilization of PEs when division is required as well 1 . In short, our novel contributions are: [15], each proposed PE of BioCare supports functionand precision-adaptability in a small area-footprint.…”