Apoth3osis
<_RESEARCH/PROJECTS

Speaking to Silicon

VERIFIED0 SORRY5 MENTAT CERTSLean 4 + Mathlib
>_VERIFICATION.SEAL
FORMALLY VERIFIED • LEAN 4 • MACHINE-CHECKED • APOTH3OSIS¬QED0 SORRY

Formal Verification Certificate

All theorems formally verified in Lean 4 with zero sorry gaps.

0 SORRY

Speaking to Silicon • Lean 4 + Mathlib • Apoth3osis Labs

The Central Question

How much does your hardware leak? Every computation leaves physical traces — execution time varies with secret data, power consumption correlates with register values, electromagnetic fields radiate internal state. This project formalizes information-theoretic side-channel security in Lean 4: leakage is mutual information I(Secret; Observable) over finite probability spaces. A device is secure when this quantity is bounded. The data processing inequality proves that post-processing observables cannot increase leakage, and composition theorems handle multiple observations.

Side-Channel Leakage Model

ChannelPhysical ObservableLeakage Measure
TimingExecution durationI(key; cycles)
PowerCurrent drawI(key; trace)
EMElectromagnetic fieldI(key; spectrum)

Key Verified Results

mutual_info_nonneg

Mutual information is non-negative

Entropy.lean

data_processing_ineq

Post-processing cannot increase leakage

Leakage.lean

composition_bound

Multi-observation leakage bounded by sum

Composition.lean

constant_time_zero_leak

Constant-time execution implies zero timing leakage

Leakage.lean