SEMRON, founded in 2020 in Dresden, develops memcapacitive compute-in-memory AI accelerator hardware and model deployment tools for continuous on-device inference on smartphones, wearables and headsets.
SEMRON is a semiconductor company founded in 2020 and based in Dresden, Germany. It develops compute-in-memory (CIM) hardware intended to run AI inference continuously and directly on consumer devices - smartphones, wearables and headsets - rather than in the cloud. The company's stated aim is to make on-device AI practical at the power and cost levels those form factors require.
Its core technology, CapRAM, is a 3D-scalable compute-in-memory approach that stores AI model weights in variable capacitance, a memcapacitive method distinct from conventional memory-based designs. SEMRON positions the technique as a route to high energy efficiency and parameter density, and argues that 3D integration can reduce the cost of the resulting AI chip. Technical work spans memcapacitive computing, AI accelerator design, 3D chip integration and edge inference.
Alongside the hardware, SEMRON provides an end-to-end deployment workflow for model developers. Models sourced from Hugging Face or supplied as custom ONNX files are compiled into efficient integer versions and packaged to run on SEMRON hardware, covering compilation and quantization alongside the accelerator itself.
The company targets consumer electronics and edge AI applications, including smartphones, wearables and AR/VR headsets.
