Public evidence

Silicon verification findings

We run our AI verification flow on respected open-source silicon — digital RTL and analog/mixed-signal netlists — and publish what holds up. Every finding names the exact commit and gives you a reproduction you can check on the public code yourself — no trust in us required. Each one survived two independent adversarial refutations before it was published, and every report also states what it cleared, so a finding list is never read without the checks it passed.

3 reports · 6 confirmed findings.

SERV — bit-serial RISC-V CPU

Digital RTL2 findings@ 41e8aee
The award-winning bit-serial RV32I RISC-V core, widely used and RISC-V-compliance tested for the M, C and Zicsr extensions.
  • Correctness bugMRET discards mepc[1]: trap-return lands two bytes early with the C extension
  • Correctness riskAn interrupt preempting MRET or a mepc/mtval CSR read vectors to MTVAL instead of MTVEC
Read the report

lulu9312 Bandgap_Reference — sky130 self-biased BGR

Analog / mixed-signal2 findings@ 5f958bc
An open-source sky130 bandgap voltage reference: a self-biased current-mirror core (NMOS mirror XM1/XM2, PMOS mirror XM3/XM4/XM5) develops a PTAT current across R1 from a 2:1 BJT ratio and sums it with a CTAT Vbe to make ~1.2 V. The README states the output is “constant … irrespective of temperature and supply variations.”
  • Design defectSelf-biased reference with no startup circuit — power-up relies on subthreshold leakage
  • Design defectOutput tracks the supply at ~85 mV/V — the “reference” does not reject VDD
Read the report

Mabrains Analog_blocks — sky130 bandgap / LDO / OTA

Analog / mixed-signal2 findings@ d67887a
Open-source sky130 analog IP from an analog design-services firm, carried through to layout with DRC/LVS scripts and datasheets: Banba-style bandgap references (1.8 V, 5 V, self-biased), Miller and folded-cascode LDOs, and Miller / folded-cascode OTAs. References and LDOs are power-management IP — the class a company like TI ships by the thousand.
  • Design defectA duplicated port name ties the supply to ground: the pre-layout bandgap testbench measures a dead circuit
  • Design riskThe 1.8 V bandgap’s OTA repeats a port name across eight files — the ground connection is silently dropped
Read the report
Want this run on your design?
The same flow, on your RTL or analog netlists, with the bugs and reproductions delivered to you — not published.

A paid pilot hunt returns confirmed findings on a module of your choosing, each reproducible on your own code within an afternoon. Pay for proof, not promises.

Start a pilot
Silicon verification findings — real bugs in open-source RTL and analog IP — ActGen