Edge AI NPU IP · Process-Independent · 55 nm to 2 nm

Edge AI NPU IP

Hand-Crafted for SoC Integration

From speech recognition to conversational LLM to computer vision — our NPU IP powers on-device Generative AI, built by an IC design team focused on hand-tuned, integration-ready IP.

~1 TOPS Gen AI Edition Whisper · Qwen · VITS Validated 55 nm → 2 nm Process-Independent
HonChen NPU
ENGINECNN + Transformer
CTRLRISC-V 5-stage
BUSAXI Master
BUSAPB · Sync FIFO
MEMLocal SRAM
RUNTIMEMulti-Model
● Products

Two NPU IP Editions

Hand-tuned RTL delivered as Protected Code. Process-flexible across TSMC, UMC, SMIC.

CNN Edition
~0.5 TOPS INT8 ~3.3 mm² @ 28nm

Vision-first lightweight NPU. Ideal for cost-sensitive edge devices.

Learn more →
Generative AI Edition
~1 TOPS INT8 + FP16 ~6.1 mm² @ 28nm

Conversational AI on edge. CNN + Transformer dual-engine for full speech / LLM / TTS.

Learn more →
● Why HonChen

Hand-Tuned RTL,
Built for Real Silicon

Hand-tuned RTL, workload-validated, process-flexible.

01

Hand-Tuned RTL

Every line of RTL hand-written by experienced RTL engineers. No HLS shortcuts. Smaller die, higher clock, better PPA.

02

Workload-Validated

Whisper Small/tiny, Qwen 1.7B/500M, VITS TTS, YOLOv3-tiny, YOLOv4-tiny — verified on our NPU IP.

03

Process-Flexible

RTL IP — not tied to any foundry or process node. From 55 nm to 2 nm class, customer chooses based on budget and product positioning.

Ready to evaluate HonChen NPU IP?

For licensing, technical evaluation, or joint development — let's talk.