SEMRON

What they do

SEMRON is designing an AI inference chip that computes inside its memory. Instead of shuttling neural network weights back and forth between memory and processor, its CapRAM technology stores weights as variable capacitance in devices called memcapacitors and performs the multiply-and-add operations of AI directly in those arrays. The arrays are planar cross-point structures that can be stacked vertically; eeNews Europe reported the company envisaging up to 100 layers, and an EU project describes a 3D NAND-style manufacturing route.

Why it matters

Running large language models on a phone or VR headset is limited by heat and battery life. SEMRON's goal, as described by SPRIND, is to run very large models continuously on chips drawing only a few hundred milliwatts. Capacitive read-out is claimed to produce less noise than resistive memory approaches, which is a known weakness of analogue in-memory computing. These are design claims; the company was building a demonstrator chip as of 2025 and 2026, so performance at product level is not yet public.

Products to know

• CapRAM: the capacitive in-memory computing core.
• Aloe AI: an EU-funded project (EIC, about 2.5 million euros, January 2025 to December 2026) coordinated by SEMRON to build a monolithic 3D-stacked inference chip demonstrator.

The German angle

SEMRON was founded in 2020 in Dresden by Aron Kirschen (CEO) and Kai-Uwe Demasius (CTO). It took part in SPRIND's New Computing Concepts challenge in 2023, and SPRIND, the federal agency for breakthrough innovation based in Leipzig, contributed to its 7.3 million euro seed round in early 2024, led by Join Capital with SquareOne, OTB Ventures and Onsight Ventures. Dresden gives it something few chip start-ups have: fabs, process engineers and equipment suppliers within a short drive.

Sources

• SPRIND: SEMRON secures seed financing
• eeNews Europe: Semron raises funds for memcapacitor processing
• CORDIS: Aloe AI project