Weebit Nano launches Matrix research program to test AI computing inside memory

Weebit Nano launched its Weebit Matrix In-Memory Compute Research Program on November 4, 2026, giving academic and industry researchers access to hardware built around its resistive random-access memory, or ReRAM. The program will let teams prototype AI algorithms, test ways to program the memory and explore architectures that perform some calculations inside the memory array. Participants will share their ReRAM findings and feedback with Weebit.

The program targets a bottleneck in AI inference, the process of running a trained model on new inputs. Conventional accelerators move model weights and other data between memory and processing units. Those transfers consume power and can delay results, particularly when memory bandwidth limits how quickly a processor can get the data it needs. In a ReRAM crossbar, the cells that store neural network weights can also take part in analog calculations. The array can perform many multiply-accumulate operations in parallel, reducing the need to move weights for each calculation.

What researchers get

Weebit is offering two packages. The Evaluation Package includes a board that gives researchers direct access to the hardware. The Quick-Start Package pairs an in-memory computing board with a field-programmable gate array (FPGA) subsystem and sample code. Both packages include packaged chips, documentation and a socketed board. The chip’s 32×32 crossbar supports experiments in vector-matrix multiplication, convolution and spiking neural networks, which use discrete pulses to carry information.

CEO Coby Hanoch traced the program back to work with CEA-Leti, including a ReRAM-based spiking neural network demonstration in 2019. He said companies have expressed interest in participating and that technical challenges remain before a commercial in-memory computing product can reach the market. Neuromorphic AI researcher Kevin D. Johnson said he plans to use Matrix to check his simulations against hardware and measure accuracy, latency and power at the system level. His work has explored applications including financial risk analysis and detection of GPS signal manipulation.

Weebit says qualified researchers can apply with a project proposal. The company will assess applications against the program’s research focus. It also plans to demonstrate Matrix at electronica 2026 in Hall C2, stand 637. Apply to the research program

The clearest potential use is in devices that run AI locally under tight power and heat limits. A battery-powered camera, for example, may need to classify images throughout the day. If its hardware spends less energy fetching model weights and can still meet the application’s accuracy and response-time requirements, the device could do more work on the same battery. Weebit also points to radiation tolerance as a reason to investigate ReRAM for aerospace and other harsh environments.

One study offers a sense of the potential gains. A U.S. developer of radiation-tolerant AI hardware evaluated Weebit ReRAM for an image-classification task. Using measurements from a ReRAM array and application-level modeling, the study projected that the proposed implementation would use about one-eightieth the power of an NVIDIA GeForce RTX 4090 implementation. It also projected inference latency below one millisecond, down from tens of milliseconds. The figures apply to that specific model, workload and hardware comparison.

Bottom line: The research kits give teams a way to test how much of that advantage survives in a complete system. An array can carry out its share of the math efficiently while the surrounding circuits still draw power to prepare inputs, digitize outputs or handle operations that run elsewhere. Researchers can also examine how analog behavior affects model accuracy and how often a system needs to recalibrate. Those trade-offs may look different for a simple image classifier and a model that needs several arrays working together. Measuring power, latency and accuracy on the same workload will help establish where ReRAM makes sense for edge AI and what needs to improve before it can support larger models.

Contact Weebit Nano

Also Read:

Weebit Nano’s FY26 ESG Performance: Technical Progress and Strategic Importance

Weebit Nano ReRAM Reaches Commercial Tape-Out Milestone

2026 Outlook with Coby Hanoch of Weebit Nano