Roadmap
From RTL to production. Six milestones — FPGA-verified or in preparation.
R&D + RTL
Architecture design, first tile implementation, simulation verification. 21 hardware modules, ~4,000 lines of Verilog.
FPGA verification
DMA data path bit-exact; TinyLlama-1.1B generates tokens with the NPU in the data path. All five core innovations verified in our own hardware (9/9 correctness tests). TRL 5 confirmed.
Scaling & firmware
Second, larger FPGA platform; up to three tiles in parallel verified in hardware, bit-exact; 32×32 compute array; our own RISC-V sequencer (new operations as firmware); our own Linux image with driver stack; 92% code coverage over the core RTL. Next: sensor benchmarks on real signal loads.
ASIC stage
Foundry partnerships being evaluated. First engineering samples for validation.
Engineering samples
Bring-up on real silicon hardware, architecture validation, first license pilot agreements with industry partners.
Production & Vexon-2
Volume licensees, larger tile arrays, higher frequency. First steps toward the long-term dream: our own Agentic AI Assistant on NeuroVexon hardware.
Why this roadmap is realistic
We don't publish promises we can't keep. What's marked "completed" is hardware-verified and reproducible. What's marked "in preparation" is being actively prepared with partners. What's marked "vision" is long-term — but consistent.