ZEN SEMICONDUCTOR
Datapath & Power

Minimum-Energy Datapath Optimization

Determines the exact minimum precision every arithmetic operation in your algorithm needs — with formal correctness guarantees — and rebuilds the datapath around it. Typically more than 90% less area and energy than a naïve fp32 implementation.

Affine-arithmetic proofs>90% area/energyPer-algorithm
Datasheet · coming soonWatch demo
Method

Prove the bound, then shrink.

Provably-minimal datapath Full-precision opbaseline Affine analysisproves error bound Reduced datapathfewer bits · min energy every reduced-precision operation carries a proven error bound — minimal, not just smaller
Affine-arithmetic analysis bounds the error before any precision is cut — so the result is provably correct, not just cheaper.
Features

Provably minimal, not just smaller.

Formal

Proven bounds

Affine-arithmetic analysis proves the error bound of every reduced-precision operation.

Efficient

>90% reduction

Massive area and energy savings versus uniform fp32.

Targeted

Per-algorithm

Tuned to the exact numerics your workload runs.

Back to Products