EDForge
Abstract
A hardware generator for Ed25519 signing and verification that predicts resource utilization, latency, and throughput across candidate architectures, so a design can be chosen to fit its deployment constraints. Manuscript in preparation.
§What it does
EDForge generates Ed25519 signing and verification hardware for FPGAs. Instead of one fixed design, it explores a space of architectures and predicts what each will cost before synthesis: resources, latency, and throughput.
- Signing and verification datapaths, designed and implemented independently.
- Resource–performance prediction models, with typical resource-category prediction errors of 5–10% against synthesized AMD Alveo U280 designs.
- A design-space exploration methodology that finds Pareto-optimal configurations within the explored space.
§Numbers
Unified signing and verification hardware, measured on a physical U280 versus simulation:
| signatures/s | verifications/s | resources | |
|---|---|---|---|
| U280 (measured) | \approx 6.0\times 10^{6} | \approx 1.0\times 10^{6} | < \frac{1}{4} |
| simulation | 7.88\times 10^{6} | 1.2\times 10^{6} | \approx \frac{1}{3} |
§Next
Verilog-level formal verification, to establish correctness guarantees for the generated hardware.