Karan Diwan

High-Performance Multipliers

32-Bit Wallace Tree Multiplication
32-Bit Wallace Tree Multiplication.

This project extends the same question from the adder study to multiplication: four 32-bit architectures, Shift-and-Add (SHA), Braun Array (BAM), Wallace Tree (WTM), and Dadda Tree (DTM), characterized on identical terms using the Sky130HD standard-cell library. It also measures how the choice of final carry-propagate adder changes overall multiplier performance, reusing the adder IP from the project above.

Post-synthesis characterization, Sky130HD

All four results below use a 64-bit Kogge-Stone Adder as the final carry-propagate stage; a Ripple-Carry variant was also evaluated to isolate the effect of that final stage.

ArchitectureAreaFmaxCrit. pathPowerADPPDP
Shift-and-Add (SHA)9,266.4 µm²~161.3 MHz6.20 ns807 µW57,4505,003
Braun Array (BAM)36,198.5 µm²~29.17 MHz34.28 ns523 mW1,240,88317,928,440
Wallace Tree (WTM)40,904.2 µm²~73.20 MHz13.66 ns159 mW558,7522,171,940
Dadda Tree (DTM)40,322.4 µm²~176.99 MHz5.65 ns154 mW227,822870,100

Latency and throughput

ArchitectureCycles / multiplicationLatencyMultiplications / second
Shift-and-Add32198.4 ns~5.04 million
Braun Array134.28 ns~29.17 million
Wallace Tree113.66 ns~73.20 million
Dadda Tree15.65 ns~176.99 million

Relative to Shift-and-Add

ArchitectureAreaFmaxPowerLatencyThroughput
Braun Array3.91×0.18×648.08×0.17×5.79×
Wallace Tree4.41×0.45×197.03×0.069×14.52×
Dadda Tree4.35×1.10×190.83×0.028×35.12×
Toolchain: Verilog, Icarus Verilog, Yosys, OpenSTA, GTKWave, Sky130HD
View repository on GitHub →
← High-Performance Adder Architectures Next: Hopfield Network →