A merged synthesis technique for fast arithmetic blocks involving sum-of-products and shifters Conference Paper uri icon

abstract

  • In modern Digital Signal Processing (DSP) and Graphics applications, the arithmetic Sum-of-Products, Shifters and Adders are important modules, contributing a significant amount to the overall delay of the system. A datapath structure consisting of multiple arithmetic sum-of-product, shifter and adder blocks is often found in the timing-critical path of the chip. In this paper, we propose a new operator-level merging technique to synthesize this type of datapath structure. In our approach, we combine the shifting operation with the partial product reduction stage of the sum-of-product blocks. This enables us to implement the functionality of the original design by using only one carry-propagate adder block (instead of two carry-propagate adders). As a result, the timing-critical path of the design gets shortened by a significant percentage and the overall performance of the design improves. Our experimental data shows that the datapath block generated by our approach is significantly faster (13.28% on average) with a modest area penalty (3.24% on average) than the corresponding block generated by a commercially available best-in-class datapath synthesis tool. These improvements were verified on placed-and-routed designs as well. 2008 IEEE.

name of conference

  • 21st International Conference on VLSI Design (VLSID 2008)

published proceedings

  • 21ST INTERNATIONAL CONFERENCE ON VLSI DESIGN: HELD JOINTLY WITH THE 7TH INTERNATIONAL CONFERENCE ON EMBEDDED SYSTEMS, PROCEEDINGS

author list (cited authors)

  • Das, S., & Khatri, S. P.

citation count

  • 0

complete list of authors

  • Das, Sabyasachi||Khatri, Sunil P

publication date

  • January 2008