Skip to content
HOMETECH DIRECTORYNEWS & DEVELOPMENTSRELOCATEEMPLOYERSNEIGHBORHOODSABOUTSchedule a Call

MemryX: The Ann Arbor Edge AI Chip Company Built by U-M Professor Wei Lu

MemryX builds the MX3 edge AI accelerator chip from Ann Arbor. Founded by U-M EECS professor Wei Lu. $63M raised including a $44M Series B in 2025.

A2Hub Editorial· Rusty Roof Media· Updated May 4, 2026

MemryX

By A2Hub Editorial. Published 2026-04-22. Last updated: 2026-04-22.

At a glance

  • Founded: 2019
  • Founder: Wei Lu, professor of Electrical Engineering and Computer Science at the University of Michigan
  • HQ: Ann Arbor, Michigan, with offices in Taipei and Hsinchu, Taiwan
  • Funding raised: Approximately $63M total, including a $44M Series B in 2025
  • Key investors: M Ventures (Merck KGaA), Invest Michigan, and additional strategic and institutional investors
  • Status: Private, growth-stage. Shipping the MX3 edge AI accelerator product line.
  • Website: memryx.com

The pitch in one line

MemryX is the Ann Arbor company building purpose-built silicon for AI inference at the edge. The MX3 accelerator runs vision and language model inference workloads on devices that cannot afford the latency, bandwidth, or power budget of cloud inference, and the company has shipped that thesis from a downtown Ann Arbor headquarters with a Taiwan engineering arm.

Founding story: from U-M memristor research to commercial silicon

MemryX was founded in 2019 by Wei Lu, a longtime EECS professor at the University of Michigan whose research has focused on novel memory devices and the architecture of computing systems built around in-memory computation. Lu's lab is one of the more prominent academic research groups working on memristor and neuromorphic computing primitives, and the intellectual foundation of MemryX runs through a decade-plus of that academic work.

The thesis that became the company is specific. Edge AI inference, meaning running AI models on devices that are not GPUs in a data center, has been growing as a workload class for the past five years. The growth has been throttled by the lack of silicon designed for the actual computational profile of inference. Most edge AI silicon either sacrifices model accuracy to fit a power budget, or sacrifices model size to fit a memory budget, or sacrifices throughput to fit a chip area budget. MemryX's bet is that a chip architecture purpose-built for inference (not training, not general-purpose computation) can win on all three vectors simultaneously, particularly for vision models in industrial, automotive, smart city, and consumer device applications.

The company set up headquarters in Ann Arbor and built a parallel engineering footprint in Taipei and Hsinchu, Taiwan, the global heartland of contract semiconductor manufacturing. That dual footprint has been the operational backbone of the company.

What they build

MemryX's flagship product is the MX3 edge AI accelerator. The MX3 is a chip designed to run pre-trained AI inference workloads (image classification, object detection, semantic segmentation, increasingly small language models) on edge devices with low power and predictable latency.

The product story spans:

The MX3 chip. Multiple chips can be cascaded on a single board to scale inference throughput linearly with model size. That cascading architecture is one of MemryX's distinguishing technical bets; it lets a single deployment run larger models than a single chip could host.

Module and board form factors. MemryX ships the silicon as part of accelerator modules and reference boards, which makes adoption easier for OEM customers who need a working hardware-and-software stack rather than a bare die.

Software toolchain. A model compilation and deployment toolchain that takes models trained in standard frameworks (PyTorch, TensorFlow, ONNX) and compiles them down to MX3 execution. The toolchain story is critical; edge AI silicon lives or dies on whether developers can deploy their models without rewriting them.

Reference applications. Vision-heavy industrial inspection, smart camera, automotive perception, and increasingly some on-device language model use cases.

The customer story is still developing. MemryX has design wins across industrial, automotive, and smart device customers, with a public-facing product that is shipping in volume. The competitive landscape includes NVIDIA Jetson at the high end, Hailo and Sima.ai in the dedicated edge inference category, and a long tail of FPGA and microcontroller-class solutions at the low end.

Funding history

MemryX has raised approximately $63M total to date, with the most recent reported milestone being a $44M Series B in 2025. The investor base spans strategic semiconductor investors (M Ventures, the corporate venture arm of Merck KGaA), Michigan-focused state and regional capital (Invest Michigan), and additional institutional and strategic investors.

The capital strategy is consistent with what semiconductor companies need: enough runway to design, fabricate, and qualify silicon, plus enough operating capital to support customer adoption cycles that run measured in years rather than quarters. The Series B sized at $44M reflects the seriousness of MemryX's position in the edge AI silicon competition; that is enough capital to ship a follow-on product generation while supporting current customer ramps.

Ann Arbor footprint

MemryX is one of the most consequential semiconductor companies operating from Ann Arbor in the modern era. The headquarters in Ann Arbor handles strategy, customer engineering, and a meaningful share of the senior engineering team. The Taipei and Hsinchu offices handle the deep silicon engineering, foundry interface, and Asia-Pacific customer engagement.

The Ann Arbor team's recruiting profile favors senior chip architects, software engineers (compiler, runtime, drivers), and customer engineering staff. The U-M EECS connection is direct; Wei Lu's academic affiliation and the lab's research lineage have made U-M a natural recruiting source.

For an Ann Arbor engineer interested in working on actual silicon rather than software, MemryX is one of a very small number of options in the city. The other meaningful semiconductor employer is KLA's North Campus Research Complex office, which is process control rather than chip design. Movellus is the other Ann Arbor / U-M semiconductor IP company. Together they form a small but meaningful Ann Arbor chip cluster.

Leadership and people

Wei Lu remains the founder figure and the technical face of the company. The operating leadership team includes semiconductor industry executives with backgrounds across silicon design, compiler engineering, and AI infrastructure go-to-market. The Taiwan operations are critical to the company's execution; the Taipei and Hsinchu engineering benches are part of why MemryX has been able to ship silicon at the cadence it has.

Why it matters

Three reasons MemryX matters in the Ann Arbor story.

First, it is one of the most credible AI hardware companies operating from a non-coastal US city. The edge AI silicon competition is intense and most of the credible competitors are concentrated in California and Israel. MemryX's Ann Arbor / Taiwan footprint is an unusual operating model for a semiconductor company at this stage.

Second, the U-M EECS lineage is direct. Wei Lu's lab is one of the highest-profile academic groups working on the architecture of in-memory and neuromorphic computing. MemryX is the commercialization of that research. That maps onto the University of Michigan tech startups story directly.

Third, MemryX sits inside Ann Arbor's Brookings AI Star Hub designation as a hardware-AI exemplar. Most of the AI companies in the city build software. MemryX builds silicon. That diversifies the city's claim to AI hub status from "AI software" toward the broader "AI infrastructure" category, which is harder to credibly claim and more durable when claimed.

Related

  • Movellus is the U-M-spinout semiconductor IP company that operates the only other prominent Ann Arbor chip-design footprint.
  • Sight Machine is the Ann Arbor manufacturing AI platform whose customers overlap with the kind of industrial vision use cases MemryX silicon targets.
  • Voxel51 is the Ann Arbor computer vision data infrastructure firm whose dataset platform is used by teams deploying the kind of models MemryX silicon runs.
  • The semiconductors category hub covers the full Ann Arbor chip bench.
  • The companies directory is the master index.

FAQ

Where is MemryX headquartered? MemryX is headquartered in Ann Arbor, Michigan, with engineering offices in Taipei and Hsinchu, Taiwan.

Who founded MemryX? Wei Lu, professor of Electrical Engineering and Computer Science at the University of Michigan, founded MemryX in 2019. The company commercializes a research lineage from Lu's lab on novel memory devices and computing architectures.

What does MemryX do? MemryX designs and sells edge AI inference accelerator chips. The flagship MX3 product runs pre-trained AI models (vision, increasingly small language models) on edge devices with low power and predictable latency. The product ships as silicon, accelerator modules, and reference boards with a model compilation toolchain.

How much funding has MemryX raised? Approximately $63M total to date, including a $44M Series B in 2025. Investors include M Ventures, Invest Michigan, and additional strategic and institutional investors.

Who are MemryX's customers? The customer base spans industrial inspection, smart camera, automotive perception, and other edge AI deployments. Specific customer references vary by deployment confidentiality.

How does MemryX compare to NVIDIA Jetson or Hailo? NVIDIA Jetson is the dominant high-end edge AI module family with a strong software ecosystem. Hailo, Sima.ai, and other dedicated edge AI silicon firms compete in similar vision-inference applications. MemryX's distinguishing technical bet is its cascading chip architecture, which lets multiple MX3 chips combine to run larger models than a single chip could host, and the in-memory computing lineage from Wei Lu's research.