systems / SystemVerilog & differential RTL verification
Selkov + Goldilocks RTL
Synthesizable SystemVerilog cores for fixed-point nonlinear glycolytic dynamics and divider-free 64-bit Goldilocks prime-field arithmetic (p = 2⁶⁴ − 2³² + 1), checked cycle-by-cycle against independent software references.
Question
How cleanly can nonlinear dynamical stepping (Selkov reaction-diffusion / limit-cycle dynamics) and ZK-friendly 64-bit Goldilocks modular reduction be expressed in divider-free RTL without hidden rounding drift or carry-chain bugs?
Method
Implements synthesizable SystemVerilog arithmetic datapaths exploiting the sparse Goldilocks identity 2⁶⁴ ≡ 2³² − 1 (mod p) alongside deterministic fixed-point state updates, driven by Icarus Verilog simulation, Python cocotb randomized differential testbenches, and Yosys synthesis checks.
Evidence
Bit-exact agreement between hardware simulation traces and independent arbitrary-precision Python reference models across edge residues (0, 1, p − 1, 2⁶⁴ − 1) and multi-step trajectories.
Limits
Simulation- and synthesis-verified RTL building blocks (`k-luecke/medals`); not taped-out silicon or a full multi-core ZK prover accelerator.