German semiconductor IP · Böblingen, Germany

Sovereign AI starts at the silicon.

Vexon-1 is a licensable NPU chip design from Germany: a proprietary compute architecture with a RISC-V control core that runs AI directly on the device — from sensors to language models, without the cloud. Three core mechanisms are registered as utility models; the architecture is proven in hardware on two FPGA platforms.

★ Supernova Finalist (Top 10) · GITEX Europe 2026
Rendering of the Vexon-1 chip design – visualization, not a manufacturing image
Visualization of the chip design – not a manufacturing image
5
mechanisms proven in hardware
2
FPGA platforms, bit-exact validated
3
utility models registered (DPMA)
100 %
in-house RTL, auditable

Four quiet weeks — because we were building

In the update video: Vexon-1 now computes as a cluster — multiple NPU tiles in parallel on one FPGA, validated against the reference model. A design that scales from sensor to production line.

Read the report →

Five core mechanisms

All five measured on FPGA hardware, reproducible against the reference model.

  • Sparse Engine

    Detects null inputs in language models and skips the multiplication. Less power, less latency, same math.

  • Adaptive Precision

    Automatically switches precision adaptively when data allows — up to 4× more throughput per cycle.

  • Multi-Tile

    Multiple compute units work in parallel — parallel throughput instead of sequential computation.

  • Hardware Attention

    The Transformer core as a dedicated state machine in hardware — including softmax and zero-score filter.

  • On-chip KV-Cache

    The language model memory lives directly on the chip instead of slow external memory. No cloud, no DRAM round-trip.

Architecture in detail →

How Vexon-1 works

The video explains in one minute how Vexon-1 computes: the 16×16 compute array with 256 multiply units, the five hardware mechanisms, and the proof on FPGA.

Discover Vexon-1 →

A licensing company for sovereign Edge-AI

NeuroVexon is a licensing company: partners integrate the Vexon IP into their own chips — the business model of the established IP houses, applied to edge AI. Three usage paths, three registered utility models as the foundation.

Three ways to use it:

  1. 1.IP Licensing (primary)
  2. 2.Chip sales (secondary)
  3. 3.Appliance / OEM module (tertiary)
IP & licensing in detail →

Right where the data is created.

From the field to the factory floor to the control room: AI where the data originates — with response times no cloud connection can match.

Explore applications
Serkan Bicen — Founder & Managing Director

Who is behind it

Serkan Bicen · Founder & Managing Director, with co-founder Cigdem Bicen

Demo, deep technical discussion or partnership — you reach the founding team directly.