---
title: ETH Zurich spin-out raises $3.8m to build edge AI chips for AR glasses
description: Zurich semiconductor startup Mosaic SoC has raised $3.8 million in pre-seed funding led by Founderful to build low-power perception chips for AR glasses and wearable devices.
author: Darie Nani (Editor-in-Chief)
date: 2026-04-30T11:47:22.457Z
updated: 2026-05-20T10:17:38.949Z
canonical: https://www.sovereignmagazine.com/article/mosaic-soc-edge-ai-chip-ar-glasses
image: https://cdn.nanimediahouse.com/mosaic-soc-founders-team.webp
categories: Artificial Intelligence, Startups, Science & Tech
content_type: Spotlight
region: Zurich
publication: Sovereign Magazine
about:
  - type: Organization
    name: Mosaic SoC
    description: Zurich-based semiconductor startup building dedicated perception chips that bring spatial intelligence to energy-constrained consumer devices such as AR glasses and smartphones. Founded in 2024 by Alfio Di Mauro and Moritz Scherer, both ETH Zurich PhDs.
    url: https://www.mosaic-soc.com
    foundingDate: 2024-04-24T00:00:00.000Z
    industry: Semiconductors
    sameAs:
      - https://www.linkedin.com/company/mosaic-soc
      - https://www.crunchbase.com/organization/mosaic-soc
---

Mosaic SoC, a Zurich-based semiconductor company building dedicated perception processors, has raised $3.8 million in a pre-seed round led by Founderful, with participation from Kick Foundation.

The company is developing integrated circuits that give consumer devices, particularly AR glasses and smartphones, the ability to map their surroundings in real time while drawing minimal power. Both co-founders hold PhDs from ETH Zurich's Integrated Systems Laboratory, where they spent years designing the kind of ultra-low-power multi-core chips that Mosaic SoC now intends to commercialise.

## Why AR glasses need a dedicated edge AI chip

The central constraint in wearable spatial computing is power. Devices such as AR glasses need to process camera and sensor data continuously to track the wearer's position, recognise objects and build local maps of their environment. Today, that workload typically falls to application processors or GPUs borrowed from smartphone architectures, which consume too much energy for a device that must run all day on a small battery.

The result is a series of compromises. Existing AR headsets tend to be heavy, warm and short-lived between charges. Making smart glasses that look and feel like ordinary glasses, rather than a computing platform worn on the face, requires perception hardware purpose-built for the task.

Mosaic SoC's approach is a dedicated co-processor that handles spatial awareness workloads independently. Rather than asking a general-purpose chip to do perception as a side job, the company's silicon is designed to do nothing else, at a fraction of the power.

## How Mosaic SoC's low power AI chip architecture works

Where most edge AI processors use one or two ARM-based cores, Mosaic SoC has built a proprietary multi-core architecture using eight or more cores, based on the open-source RISC-V instruction set. The design includes a memory architecture that allows parallel read-and-write access across multiple accelerators simultaneously, eliminating the bottleneck that serialises memory access in conventional designs.

The chip handles workloads such as simultaneous localisation and mapping (SLAM), eye tracking and hand tracking. In practical terms, it can build a local map of a room, track objects within it and, in a smartphone application, act as a co-processor for the front camera, running always-on classification at low enough power that the battery impact is negligible.

The company ships the chip with a full software stack, including perception algorithms and neural inference tools, which it develops and maintains. For the original design manufacturers (ODMs) building AR glasses and mobile devices, the pitch is that integrating Mosaic SoC's chip removes engineering complexity rather than adding it.

## The PULP platform and the founders' research background

Both co-founders, Alfio Di Mauro (CEO) and Moritz Scherer (CTO), trained under Professor Luca Benini at [ETH Zurich](https://www.sovereignmagazine.com/article/the-swiss-startup-cooling-down-ai-s-21-billion-overheating-problem)'s Integrated Systems Laboratory. Benini directs the PULP (Parallel Ultra Low Power) platform, a collaboration between ETH Zurich and the University of Bologna that has produced more than 40 chip designs since 2013 and won the 2024 TCMM Open Source Hardware Contribution Award. Benini is listed as an adviser to Mosaic SoC.

Di Mauro's research focused on neuromorphic computing and event-based processing, with publications cited over 1,000 times. His most-cited work includes Vega, a ten-core system-on-chip for IoT endpoints with neural network acceleration. Scherer's work concentrated on deploying machine learning models on severely resource-constrained hardware, including gesture recognition systems for short-range radar and neural network inference accelerators for nano-drones.

The academic lineage is relevant because Mosaic SoC's multi-core RISC-V architecture descends directly from the PULP platform. The founders are commercialising research they built, not adapting someone else's.

## Which companies make edge AI processors for wearable devices

The competitive landscape has shifted notably in the past two years. Intel shut down its Movidius visual processing unit line, leaving a gap in dedicated edge vision silicon. Hailo, the Israeli [edge AI chip company](https://www.sovereignmagazine.com/article/the-ai-chip-wars-heat-up-how-openai-and-broadcom-are-reshaping-manufacturing-s-future) that had raised $564 million, saw its valuation halved and cut roughly 10 per cent of its workforce in early 2026, refocusing on robotics rather than consumer wearables.

Axelera AI, based in the Netherlands, raised more than $250 million in February 2026 but targets industrial edge computing and data centre inference at power levels well above what a pair of glasses can accommodate. Syntiant, a US company that ships neural decision processors at sub-milliwatt power levels, is the closest comparable, though its primary focus has been audio processing and wake-word detection rather than spatial perception.

Qualcomm's Snapdragon AR2 platform remains the dominant choice for premium XR headsets, but it operates at a significantly higher power envelope than what ultra-light consumer glasses require. The space between what platform chips offer and what a form factor resembling ordinary spectacles demands is where Mosaic SoC has positioned itself.

## Early revenue and the chip startup business model

Mosaic SoC has already generated revenue through non-recurring engineering contracts with ODM partners, an unusual position for a company at the pre-seed stage, particularly in semiconductors where the path from founding to first revenue is typically measured in years. The company expects its revenue to shift from engineering services toward scalable chip sales as its products reach market.

The $3.8 million round is modest by semiconductor standards, where a single tape-out can cost several million dollars. The company's use of RISC-V, and its ability to draw on open-source IP from the PULP platform, likely gives it a cost advantage in development that a company building on proprietary ARM architectures would not have. The team currently numbers six people.

## Who invested in Mosaic SoC's pre-seed round

Founderful, which led the round, is Switzerland's largest [pre-seed fund](https://www.sovereignmagazine.com/article/avian-raises-2-6m-ai-thermal-monitoring). The firm, previously known as Wingman Ventures before rebranding in 2024, closed its second fund at $140 million in November 2024. It was founded by Pascal Mathis, co-founder of GetYourGuide, and typically writes cheques of $1 million to $2 million into [Swiss technology spin-outs](https://www.sovereignmagazine.com/article/swiss-startup-chipmind-says-europe-s-43bn-chip-bet-missed-the-point). Its portfolio companies have collectively raised more than $450 million in follow-on funding.

Kick Foundation, which also participated, operates the Venture Kick programme that has supported nearly 1,000 Swiss startups since 2007. Mosaic SoC received CHF 150,000 through Venture Kick in 2025. The foundation's related investment vehicle, Kickfund Ventures, raised CHF 70 million for co-investments in Swiss deep-tech companies.

Mosaic SoC was named in the TOP 100 Swiss Startups 2025 ranking and represented Switzerland at Mobile World Congress in Barcelona in March 2026 as part of the Venture Leaders Mobile programme.

**About Mosaic SoC**

Zurich-based semiconductor startup building dedicated perception chips that bring spatial intelligence to energy-constrained consumer devices such as AR glasses and smartphones. Founded in 2024 by Alfio Di Mauro and Moritz Scherer, both ETH Zurich PhDs.

[Website](https://www.mosaic-soc.com)

## FAQ

**Q: What is an edge AI chip and how does it differ from a cloud AI processor?**
An edge AI chip processes artificial intelligence workloads directly on the device, rather than sending data to a remote server. This reduces latency, improves privacy and avoids the power cost of continuous wireless data transmission. Mosaic SoC's chip is designed specifically for spatial perception tasks such as mapping and object recognition.

**Q: Do AI chips for AR glasses require more power than standard mobile processors?**
Current AR glasses often rely on application processors or GPUs designed for smartphones, which consume more power than a small wearable battery can sustain for a full day. Dedicated edge AI chips such as those developed by Mosaic SoC are engineered to perform perception tasks at a fraction of the power, enabling lighter devices with longer battery life.

**Q: Who are the leading companies making edge AI chips?**
The field includes Qualcomm (Snapdragon AR2 for XR headsets), Hailo (edge AI for robotics and industrial applications), Axelera AI (high-performance edge inference) and Syntiant (ultra-low-power neural processors). Mosaic SoC differentiates by targeting the sub-one-watt power class specifically for spatial perception in consumer wearables.

**Q: What processor architecture does Mosaic SoC use?**
Mosaic SoC uses a proprietary multi-core architecture based on the RISC-V instruction set, with eight or more cores. This differs from competing designs that typically use one or two ARM-based cores. The architecture descends from the PULP (Parallel Ultra Low Power) platform developed at ETH Zurich.