Trusted Compute Unit · Compute Gear

SPL-G1 硬件级因果审计 · 可信计算单元

SPL-G1 is not a general-purpose CPU/GPU. It is a dedicated security primitive — a hardware causal-audit Trusted Compute Unit (TCU) that delivers provable, hardware-level causal audit across the entire compute lifecycle. Every operation is anchored to an immutable causal chain.

GitHub Repository ↗ SourceForge
What it is

A compute unit that audits itself

Where conventional processors separate computation from trust, SPL-G1 fuses them: the audit is part of the silicon, not a layer bolted on after the fact.

A2 · NUMERIC

True IEEE-754 FP16

A real half-precision floating-point datapath (sign / 5-bit exponent / 10-bit mantissa) with subnormal, NaN and ±Inf handling — not a fixed-point approximation. Verified in RTL simulation.

A3 · COMPILER

The splcc compiler

A C-subset compiler (splcc.py v0.1) that lowers C into SPL-G1 microcode, plus splcc_bridge.py for host-side integration. High-level intent becomes auditable hardware flow.

A1 · CONTROL

Sequencer v4 + RA-BUS

A parameterized 256-entry program memory with JMP / JZ / JNZ / CALL / RET / HALT, an 8-deep return stack and out-of-bounds protection. The RA-BUS READ transaction state drives the external data channel (A4).

A6 · ANCHOR

SBC fuse & causal anchor

Each operation is sealed to a 256-bit anchor through an SBC fuse. Once fused, the causal chain is immutable — the root of provable auditability.

A5 · CONSTRAINT

Causal constraint unit (v2)

Hard-constraint checking via an all-ones constraint_bits comparison plus a 56-bit dependency mask with cascade invalidation. Pass-through mode always passes; configured rules are enforced at dispatch.

FABRIC

2D PIM array + RA-BUS

A two-dimensional in-memory-compute array coupled to a 4-target RA-BUS (compute / sequencer / external memory / anchor) — data-centric flow with no stored-program bottleneck between memory and compute.

Specification

At a glance

Classification
Trusted Compute Unit (TCU) — dedicated causal-audit security primitive
Compute fabric
2D in-memory-compute (PIM) array
Numeric core
True IEEE-754 half-precision FP16 (subnormal / NaN / ±Inf)
Sequencer
v4 — 256-entry program memory, control-flow, RA-BUS READ (v5 state)
Data channel
RA-BUS, 4 targets (compute / sequencer / ext-mem / anchor)
Compiler
splcc v0.1 (C → microcode) + splcc_bridge.py
Anchor
256-bit causal anchor, SBC fuse (immutable chain)
Audit
Constraint unit v2 — all-ones check + 56-bit dependency mask
Simulation
Icarus Verilog — 0 errors, 7/7 tests PASS (Phase A)
License
Open source — see repository LICENSE
Status

Phase A — core loop complete

Milestones A1 (control-flow), A2 (FP16), A3 (compiler), A4 (data channel) and A6 (SBC fuse) are delivered and verified in RTL simulation. A5 causal constraints are at the v2 level; the programmable per-class mask is planned.

A1 control-flow ✓ A2 FP16 ✓ A3 splcc ✓ A4 data channel ✓ A5 constraint v2 A6 SBC fuse ✓ Sim 7/7 PASS
Read the source ↗ ← Back to ecosystem