Lightweight edge chip
Forge
A complete language model inside a watt-class handheld device.

Lightweight edge chip
MaskROM compute-in-memory: Weights are physically cast into the via array at the mask stage of tape-out. Zero external weight traffic at runtime.
Specifications
- Core model
- Qwen 3.5 0.8B
- Decode speed
- 17,000 token/s (design target)
- System power
- Watt-class
- Route
- MaskROM compute-in-memory
- Model updates
- Fixed at tape-out
- Process
- SMIC 28 nm
- Package / interfaces
- Per project
- Development kit
- Evaluation board, SDK and model mapping tools
Figures on this page are current product definitions and design targets. Final specifications, test conditions, availability and delivery versions are confirmed in project documents agreed by both parties.

Typical applications
- Handheld translators
- Voice-memo hardware
- Offline in-car voice assistants
Inference runs locally and data is processed on the device; it works fully offline.
Qwen3.5 → 3.6 → 3.8
Newer Qwen models, no new architecture.
Qwen3.6 and Qwen3.8 keep the Qwen3.5 hybrid layout, which Fanshi inference cores already map. A newer model in the same family is a new chip build, not a redesign.
From an email to a chip
01
Pick the route
Target model, performance and power goals, update cadence, interfaces. The first question is always: how often does the model change?
02
Evaluate
Run the model on the evaluation board, measure accuracy and throughput, and get a working engineering sample before any silicon is committed.
03
Design in
Interfaces, system software and field validation built around the evaluation results, turning the sample into a product prototype.
04
Deliver the ASIC
Once the model is frozen: a production chip, or the IP integrated into your own silicon.
Further reading
How to Choose an Edge LLM Inference Chip: 7 QuestionsChoosing an edge LLM inference chip comes down to model fit, power budget and how the model updates, not peak TOPS. Seven questions to answer, plus a checklist.
MaskROM Compute-in-Memory: What Casting Weights in Silicon MeansMaskROM compute-in-memory writes model weights into a via array at the mask stage, so nothing moves at runtime. How it works, why it costs least, its limits.
LLM Inference Cost: API vs Cloud GPU vs Edge Chip Over 3 YearsCount LLM inference cost over three years: an API per token, a cloud GPU per hour, or an edge chip bought once. Methods, a worked example and payback.
Ask about Forge
Send the target model, performance and power goals, interfaces and timeline. We reply with a fit assessment and an evaluation plan.


