- 01NTU expert validated Chipforge on FPGA hardware, unaided.
- 02Internal 262 tests: up to 28pp ahead of GPT-5, Claude 4 Sonnet and MAGE.
- 03Pathkey targets paid APAC projects.
Pathkey.AI (ASX: PKY) has reported positive results from validation testing of its Chipforge AI-native semiconductor design and verification platform by an independent expert at Nanyang Technological University in Singapore.
Senior lecturer from the College of Computing and Data Science Dr Smitha Kavallur Pisharath Gopi used Chipforge to take written specifications through to working operation on field programmable gate array hardware without assistance from the Chipforge team and without manual edits to the generated code.
Dr Gopi entered the design problems as design intent and Chipforge planned the design, generated the chip code, built the tests, ran the simulations, revised the design if tests failed, and produced the file that programs the chip before being deployed to physical hardware where it ran as specified.
The evaluation covered the AMD Vivado workflow and confirmed delivery of a functioning platform utilising Chipforge intellectual property with core features necessary for end users to experience the product’s primary value proposition as demonstrated through real world use, testing, and validation in a working environment.
Internal Problem Testing
Chipforge was also tested internally on 262 problems across four categories on the public CVDP (comprehensive verilog design problems) benchmark designed by American technology company Nvidia.
For each problem — which included creating a chip design from written requirements, completing unfinished code, modifying an existing design and fixing bugs — the platform generated or edited the verilog chip code and submitted it to an open-source simulator which ran the benchmark’s hidden tests to check that the design behaved as required.
Each problem was considered to be solved once every test had achieved a first-pass result, with partial credits, human grading or AI self-assessment not accepted.
The platform was reported to have scored up to 28 percentage points higher than results from using GPT-5, Claude 4 Sonnet, and the MAGE AI large language models on the same problems.
Design Intent Problems
Pathkey.AI chief executive officer Andrew Farnsworth was pleased with the positive outcome of the independent review.
“Dr Gopi entered a set of design problems as design intent and through direct prompting, the Chipforge platform guided her through planning and specification, then code development and the verification cycle, including creating the test benches and running the simulations,” he said.
“From there it moved through synthesis, generated the bitstream, deployed it to the physical hardware, and ran the tests successfully and all of that happened with no intervention or support from the Chipforge team, which is very encouraging to see.”
“We are now no longer just talking about what the platform will do—we can show our customers what it actually does and we are hoping to turn that capability into paid projects across the Asia-Pacific region including Singapore, Malaysia, Vietnam, Japan, and Australia.”
Chipforge Platform Acquisition
Pathkey.AI announced the acquisition of Singapore-based platform developer Chipforge in April, extending its AI work beyond clinical trial design and drug discovery into semiconductor engineering.
The Chipforge platform translates high-level design intent into verified, synthesisable hardware code, targeting refinement of a design process that can otherwise take 12 to 24 months and cost millions of dollars.
Initial engagements are fixed-price projects, with successful delivery potentially leading to further projects, usage fees, and software licences.
The platform is currently in pre-production release and the Chipforge Studio broader product release is targeted for early in the new year.
Get the wire before the market opens.
The ASX small-cap stories that matter, filed before 9am AEST. Curated by the Small Caps desk.