Ferite

What it is

Two Rust toolchains in one workspace. Lithic builds circuits in Rust, simulates them clock by clock, emits Verilog, and checks equivalence and synthesis. Strata partitions inference graphs across backends, compiles what it can and falls back to the CPU for the rest. The loop where Strata workloads shape Lithic hardware — and that hardware becomes a Strata backend — isn't built yet, and the project says so on its front page.

My part, scoped exactly

I built the documentation site: its deployment to Cloudflare Workers, the brand system (tokens, marks, logos, social images) and a UX pass that split one 1,700-line stylesheet into a token-based system, with new figure components and accessibility and layout verification scripts.

I did not write the Rust toolchain. That's my co-maintainer's work at Ostrium Labs.

Why the docs had to be honest

The project publishes benchmarks that go against it, and the site had to present them without flinching.

Simulated cycles are converted to time at an assumed 1 GHz and compared against real Rust crates, after checking both produce identical output. The crc3 design is 2.24× faster than the crc crate and hex 1.38× faster. Software wins elsewhere: crc32fast by 8.6×, base64 by 13.8×, memchr by 4.2×.

Synthesis is reported as Yosys generic-gate counts — explicitly not an area figure. sha256 comes to 268,470 gates, aes 181,752, chacha20 166,924, crc3 52. The block ciphers come out seven to eleven times larger than the roughly 24,000-cell budget of the target die. The plate for this project draws exactly that.

Tradeoffs

• A token system costs more up front than a stylesheet. It paid for itself the first time the figures needed a dark variant.
• Verification scripts for layout and accessibility run against the built site, so they catch regressions late rather than at authoring time.

Where it stands

Pre-release, v0.0.1. Nothing published to crates.io.