Verified RTL/PPA optimization
Turn a trusted RTL baseline into a leaner, faster implementation, with correctness preserved at every step.
One owned block. Your flow. Evidence before adoption.
Three correctness-gated RTL transformations, evaluated under pinned academic VTR/PTM 45 nm contracts.
Evaluation system
Every candidate advances through the same frozen functional, formal, implementation, and evidence checks before it can become a result.
Set one block, its interface, objective, target flow, validity limits, and evidence requirements.
Reproduce the starting RTL, its checks, and its implementation measurement.
Explore structural changes while holding the evaluation contract constant.
Run functional, formal, synthesis, route, and evidence-integrity gates.
Return accepted RTL and its exact evidence package, or a bounded no-improvement finding.
Choose the opportunity
Select an implementation important enough to matter, bounded enough to measure, and owned by engineers who can evaluate the change.
From baseline to candidate
Preserve the external behavior, reorganize the internal structure, remove unnecessary cost, and prove the result before it advances.
Conceptual visualization: the module interface remains fixed while internal structures are analyzed, removed, merged, rerouted, and verified.
Shared Boolean logic and simplified state readout under a frozen command interface.
Range-correct arithmetic widths and a balanced exact-addition tree.
A different sequential state representation connected by a proved refinement relation.
These are academic VTR/PTM 45 nm post-route estimates on a homogeneous LUT6 target. They are not ASIC signoff, commercial-FPGA characterization, physical-board measurements, measured energy, or manufactured-silicon evidence. Percentages apply only to each case's frozen contract and are not a cross-circuit performance ranking. The ML-KEM CBD case is not side-channel analysis or certification of a complete cryptographic implementation.
Bring your real constraints
The public portfolio qualifies the method under a pinned academic proxy. A customer pilot replaces that proxy with customer-owned RTL, assumptions, libraries, constraints, activity, tools, and acceptance policy.
Start the evaluation
The first conversation should establish the function, target, current flow, correctness stack, primary objective, confidentiality boundary, and evaluation window.
Do not attach confidential RTL to the first message.