TechnologyWhat we build
FELSITE · 2026

Sovereign hardware roots of trust,
on open silicon

RISC-V, post-quantum cryptography, and designs built to be audited.

01 · The primer

A hardware root of trust does four things

  • Passport. Proves the device is genuine — not a clone, not a swapped board.
  • Bouncer. Checks the software before it runs. Wrong signature, the system does not start.
  • Safe. Designed to keep keys inside a protected hardware boundary, with resistance to software, fault and physical attack.
  • Notary. Signs a statement about what the machine is running, so someone far away can verify it.

Everything above it depends on it being correct — and little else in the system is positioned to check it.

02 · The open problems

The questions with no settled answer

The interesting part is not what a root of trust does. It is what nobody has agreed on:

  • How do you build a monitor that watches a system without becoming part of what must be trusted?
  • How do you compose one proof from devices with no reason to trust each other?
  • What belongs in fixed logic, and what in firmware? Tape-out freezes the answer.
  • How much die area is side-channel resistance worth?
  • How do you design something an evaluator can reason about, rather than something that passes tests?

We have opinions on all five. They are why the part looks the way it does.

03 · Foundations

What we build on

SHAKTI, the RISC-V core from IIT Madras · Tock, a memory-safe embedded OS in Rust · an open root-of-trust silicon foundation · an open post-quantum engine · ML-DSA and ML-KEM · SPDM, DICE, TPM 2.0.

Where the line sits. Open and auditable by anyone: the core, the OS, the cryptographic foundations, and every contribution we make upstream. Ours and deliberately unpublished: the product architecture, the security subsystems we add, the threat model, and the certification evidence. A closed die asks a buyer to trust a vendor; this lets them verify.

04 · Status

Verified in software. Moving to silicon.

Key security properties formally verified. Functional regressions passing in simulation and emulation. Synthesises onto the prototype platform with margin, and prototype bring-up is under way. We are pre-silicon; the detailed schedule is shared under NDA.

05 · Verifiable in public

Five patches merged upstream

Tock, the SHAKTI C-Class, the PULP platform. Commits in the open, reviewed by maintainers who owe us nothing — every row links to the upstream project’s own repository.

GitHubGitLab

Operating systems — Tock OS

ContributionUpstreamStatus
RV64 userspace CI testPR #4880MERGED
Enable switch_to_process on RV64PR #4878MERGED
running_thread_id() for RV64PR #4871MERGED
RV64 CSR compilation fixPR #4865MERGED
SHAKTI C-Class board supportA board and chip crate so Tock runs on the SHAKTI C-Class, an Indian open-source RISC-V core.PR #5051IN REVIEW · CI PASSING
64-bit syscall ABI specification (TRD)Contributed implementation findings and review that shaped the final design, including an explicit address-carrying return type.PR #4906OPEN · 2 APPROVALS

RISC-V processors — SHAKTI C-Class (IIT Madras)

ContributionUpstreamStatus
PMP specification compliance + Smepmp/ePMPMR !87OPEN
dcsr.stepie single-step defectReported with reproduction and a one-term fix; implemented upstream by a SHAKTI maintainer.Report #301 · Fix !231ADOPTED
RV64 >8 PMP entries fail to buildIssue #7OPEN
Smepmp core threadingMR !229DRAFT
Debug-subsystem findings#298 · #299 · #300OPEN
CLINT 64-bit timer readIssue #139OPEN

Hardware IP & security tooling — PULP platform (ETH Zürich / Università di Bologna), PROLEAD (Ruhr-Universität Bochum)

ContributionUpstreamStatus
AXI memory-error masking fixPR #426MERGED
RISC-V debug: halt-on-resetPR #188IN REVIEW
TileLink adapter: handshake fixPR #63OPEN
TileLink adapter: read byte-mask fixPR #64OPEN
PROLEAD build portabilityPR #18OPEN