
Multimodal fusion inference IP
FS-FusionJoint vision-language understanding (VLM) and multi-sensor fusion, computed in place at the edge.
Native SmolVLM + SegFormer MiT fusion pipeline
FS-Fusion detailsFanshi casts large language models directly into compute-in-memory silicon: three silicon routes, one family of inference cores, one line from model to tape-out. No network, no data leaving the device, near-zero marginal cost per inference.
The same inference cores and three silicon routes can cast any model into a chip. Pick one that fits, or have one built around your model.

GPUs and NPUs spend most of their time and power shuttling weights between memory and compute. Fanshi turns that around: the weights live in the memory array, and the math happens where they are.
Weights on the move, energy on the move.
External weight traffic per token
800MB
Traffic comes from
Off-chip reads · cross-chip transfer
0.8B · INT8 estimate, not system powerSeparate memory
In a conventional NPU or GPU, memory and compute are separate pieces of silicon. For every token generated, all of the weights cross the bus: bandwidth sets the ceiling, and most of the energy is spent in transit.
Compute-in-memory
Fanshi builds the multiply-accumulate circuits into the memory array itself. Weights stay in the array, activations flow in, results flow out. No weights on the bus.
MaskROM
Mask ROM compute-in-memory writes the weights physically into the via array during tape-out. The chip leaves the fab with the model already inside, so power and marginal cost per inference drop to the minimum.
All three move the weights less, or not at all. They differ in where the weights live and whether the model can still change. Each production chip takes one route; for a custom build, the route follows how often your model changes.
On-device decode speed
Forge design target · Qwen 3.5 0.8B
System power
Forge, lightweight edge class
Data stays on the device
Local inference, no network needed
Marginal cost per inference
After the one-time chip purchase
Pick by workload: a watt-class handheld, a node that runs 35B-parameter models, or a vision chip that keeps up with every frame. If none fits, we build one around your model.

A complete language model inside a watt-class handheld device.
Typical applications
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.
License Fanshi inference cores into your own ASIC or FPGA. One core family, four architectures, running on an evaluation board first.

Joint vision-language understanding (VLM) and multi-sensor fusion, computed in place at the edge.
Native SmolVLM + SegFormer MiT fusion pipeline
FS-Fusion details
Native support for language models built on hybrid attention, including the gated full-attention layers of the Qwen3.5 family.
Qwen3 / 3.5 (0.6B to 9B) and SmolLM2
FS-Attention details
Built for dynamically activated sparse models, carrying large parameter counts within an edge bandwidth budget.
Large sparse models on edge budgets
FS-MoE details
Linear attention and RNN-style architectures led by DeltaNet, including the Gated DeltaNet layers used from Qwen3.5 onwards.
Very low-latency streaming with a fixed-size state, however long the context
FS-Linear detailsThe evaluation board and customisable IP licensing validate model mapping, interfaces and the full system path. An engineering sample exists before the silicon does, so design-in and tape-out decisions rest on measurements.
Ask about IP licensing
That is how the first-generation chips were made, and yours can follow the same path: run it on an evaluation board first, then choose between an existing chip and one of your own.
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.
Hardware is a one-off purchase. A token bill never stops. Pick the model you use today and drag the daily volume.
Self-hosted cloud GPU estimated at $0.70 per million tokens (2026 median H100 rental). After the one-time chip purchase, the marginal cost per inference is ≈ 0.
Three-year inference cost
Pays back within the first month
Prices are blended 2026 public API list prices (input:output ≈ 1:1). Illustrative only; actual costs depend on usage and quotation.