A LUT-Size-Aware Design Rule for Pipelined Bit-Level Multiply-Accumulate Units in FPGA Signal Processing
Abstract Multiply-accumulate (MAC) units built in field-programmable gate array (FPGA) logic are needed for FIR filters, convolution, and matrix products when the digital signal processing (DSP) blocks of a device serve other functions, but look-up-table (LUT) multipliers limit their clock. Bit-level MAC units avoid multipliers by reading bit-plane counts from small tables, yet the table size has so far been fixed without regard to the device. This paper derives a LUT-size-aware design rule: a table that counts two G-bit groups has 2G address bits, so the group size G equal to half the number of LUT inputs makes every table out put bit one LUT. A pipeline register is placed where post-route timing analysis locates the critical path, and a signed extension reuses the same tables. On a Zynq UltraScale+ device with 6-input LUTs, the rule selects G = 3, and the resulting processing element meets a 400 MHz constraint for 16 eight-bit values, 2.2 times the clock of a LUT-multiplier element, with 11% fewer LUTs and 18% less power than its unpipelined form; its systolic arrays sustain 269–286 MHz. On a Lattice ECP5 device with 4-input LUTs, the rule correctly predicts G = 2: the pipelined element is the smallest design tested on that device and runs at 102 MHz. A 16-tap FIR filter built from one element reproduces a reference
Authors
- srinivasa rao kollimarla
- K Sana
Institutions
- Jawaharlal Nehru Technological University Anantapur (IN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-17
- DOI
- https://doi.org/10.21203/rs.3.rs-11742093/v1
- Primary Topic
- Digital Filter Design and Implementation
- Type
- preprint