PIR-DSP: An FPGA DSP Block Architecture for Multi-precision Deep Neural Networks
Authors:
- 1 April 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 35-44
Abstract
Quantisation is a key optimisation strategy to improve the performance of floating-point deep neural network (DNN) accelerators. Digital signal processing (DSP) blocks on field-programmable gate arrays are not efficiently utilised when the accelerator precision is much lower than the DSP precision. Through three modifications to Xilinx DSP48E2 DSP blocks, we address this issue for important computations in embedded DNN accelerators, namely the standard, depth-wise, and pointwise convolutional layers. First, we propose a flexible precision, run-time decomposable multiplier architecture for CNN implementations. Second, we propose a significant upgrade to DSPDSP interconnect, providing a semi-2D low precision chaining capability which supports our low-precision multiplier. Finally, we improve data reuse via a register file which can also be configured as FIFO. Compared with the 27 × 18-bit mode in the Xilinx DSP48E2, our Precision, Interconnect, and Reuseoptimised DSP (PIR-DSP) offers a 6× improvement in multiplyaccumulate operations per DSP in the 9 × 9-bit case, 12× for 4 × 4 bits, and 24× for 2 × 2 bits. We estimate that PIR-DSP decreases the run time energy to 31/19/13% of the original value in a 9/4/2-bit MobileNet-v2 DNN implementation.Keywords
This publication has 38 references indexed in Scilit:
- Scaling equations for the accurate prediction of CMOS device performance from 180 nm to 7 nmIntegration, 2017
- ESE2017
- Frequency Domain Acceleration of Convolutional Neural Networks on CPU-FPGA Shared Memory System2017
- Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural NetworksIEEE Journal of Solid-State Circuits, 2016
- XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks2016
- Going Deeper with Embedded FPGA Platform for Convolutional Neural Network2016
- A 65nm 3.2GHz 44.2mW Low-Vt register file with robust low-capacitance dynamic local bitlines2015
- SLIP2015
- Low-power level converting flip-flop with a conditional clock technique in dual supply systemsMicroelectronics Journal, 2014
- Quantifying the Gap Between FPGA and Custom CMOS to Aid Microarchitectural DesignIEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2013