Skip to content

Move less. Keep inference on the device.

Fanshi organises its technology around where model weights are stored and computed: MaskROM compute-in-memory, SRAM compute-in-memory and near-memory compute.

A 300 mm wafer on a probe station
Fig. 1 · A 300 mm wafer on a probe station

Cast the whole model
into a single chip.

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.

Inside the silicon / ARCHITECTURE STUDY03 / 03

Make the weights part of the silicon.

MaskROM · cross-section

External weight traffic per token

0MB

What remains at runtime

In-place compute · no external weight traffic

0.8B · INT8 estimate, not system power
Fig. 2 · Where the weights live decides where the power goesSchematic · not an actual layout
  1. Separate memory

    Every token moves every weight

    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.

  2. Compute-in-memory

    Put the weights in the memory array

    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.

  3. MaskROM

    Cast at tape-out, never moved at runtime

    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.

One goal,
three routes.

All three move the weights less, or not at all. They differ in where the weights live and whether the model can still change. The first question to answer: how often does your model change?

MaskROMCast in the maskMaskROM compute-in-memory
Where weights live
Mask layers, cast at tape-out
Runtime movement
None
Model updates
Fixed at tape-out
Best for
Stable models, high volume, lowest power and marginal cost
Forge
SRAM CIMCompute in SRAMSRAM compute-in-memory
Where weights live
Inside the SRAM array
Runtime movement
MAC inside the array
Model updates
Loaded at boot, updatable
Best for
Mid-size models, devices that need to swap models in the field
Argus
NMCCompute beside memoryNear-memory compute
Where weights live
High-bandwidth external memory
Runtime movement
On-chip streaming
Model updates
Continuous iteration
Best for
Large, fast-moving models in heavy-load nodes
Trident
17,000token/s

On-device decode speed

Forge design target · Qwen 3.5 0.8B

~1 W

System power

Forge, lightweight edge class

100%

Data stays on the device

Local inference, no network needed

≈ $0

Marginal cost per inference

After the one-time chip purchase

Choose the route for your project

Send the target model, performance and power goals, and how often the model changes. We reply with a route and an evaluation plan.

Request an evaluation kit