Weebit Matrix™ IMC Program
Accelerating In-Memory Compute for Next Generation AI
Get Started with Weebit Matrix™ IMC Program
In traditional computing architectures, moving data between the processor and memory creates performance and bandwidth bottlenecks, and consumes significant power. In-memory computing (IMC) addresses this constraint by performing operations directly within the memory array, minimizing data transfer and enabling higher throughput at significantly lower power.
Weebit ReRAM is well-suited for IMC implementations. It combines ultra-low power operation, BEOL integration, scalable cell architecture, and analog programmability required for efficient vector-matrix computation. Its inherent radiation tolerance also supports deployment in aerospace, defense, and medical environments.
Weebit Matrix™ IMC Development Platforms
The Weebit Matrix In-Memory Compute (IMC) program enables engineers and researchers to evaluate and develop ReRAM-based IMC architectures on silicon. It provides a structured platform for implementing, validating, and benchmarking IMC algorithms and system designs.
The platform is intended for both industry and academic users working on AI acceleration, near-memory compute, and emerging architectures. It supports hands-on exploration of ReRAM crossbar behavior. The platform can only be used for research and prototyping purposes, with no commercial use rights.
The complete development platform includes ReRAM crossbar array chips, dedicated boards, FPGA controllers, and supporting sample code and documentation.
Two platform configurations are available:
- Weebit Matrix IMC Evaluation – Provides direct analog access for device-level interaction and detailed electrical characterization of the 32×32 ReRAM crossbar array
- Weebit Matrix IMC Quick-Start – Integrated development environment for system-level work, supporting matrix-vector operations and neural network inference using the 32×32 ReRAM crossbar array, with onboard FPGA and control logic
Using the Weebit Matrix™ IMC platform, researchers and developers can:
- Apply and control voltages across word-lines, bit-lines, and select-lines
- Program and read ReRAM cell states, including multiple resistance levels
- Execute in-memory operations, including vector-matrix multiply-accumulate (MAC) and convolution
- Scale the effective array size by interconnecting multiple boards
- Perform stress and reliability testing under radiation, high-temperature, and EMI conditions
- Develop and validate algorithms and architectures, based on real silicon behavior
Key Benefits
- Performance Gains: Compute directly in memory, reducing latency and allowing significantly faster and wider parallel processing
- Power Savings: 10x-100x lower energy per inference operation
- Parallel Compute Capability: Native vector by matrix multiply-accumulate (MAC)
- Flexible Platform Options: Supports both device-level characterization and system-level development
- Complete Development Kit: Integrated delivery of chips, boards, FPGA control, and sample code
- Engineering Community Impact: Enabling universities, startups, and enterprises to prototype and publish quickly
A Broad Range of Applications
The Weebit Matrix IMC platform supports exploration and development of AI and data-centric workloads where data movement and power efficiency are critical, including:
- AI and Machine Learning (Inference): Efficient execution of neural networks using in-memory vector–matrix operations
- Edge and IoT Systems: Ultra-low-power processing for constrained devices
- Computer Vision: Object detection, image classification, and sensor-based perception
- Natural Language Processing at the Edge: Keyword spotting and compact language models
- Autonomous Systems: Drones, robotics, and industrial automation requiring real-time decision-making
- Smart Sensors: Always-on sensing for audio, vision, and environmental monitoring
- Data Processing Acceleration: Search, filtering, and pattern-matching workloads
Feature Specifications
| Feature | Description |
|---|---|
| Chip Architecture | 32×32 crossbar array (1T1R ReRAM cells) |
| Package | 128-pin QFN |
| Operating Range | Typical: 25°C, 5.5V supply |
| Array Functions | Read, Set, Reset |
| System Functions | ReRAM array address selection, power supply value setting (Input), current measurement (Output) |
| IMC Operations | Vector-matrix Multiply-and-Accumulate (MAC) |
Get Started with Weebit Matrix™ IMC Program
“The EDGE AI FOUNDATION is dedicated to building a dynamic ecosystem that makes intelligent systems more capable, efficient, and sustainable. We are excited to welcome Weebit Nano, whose ReRAM technology offers new ways to unlock performance and efficiency for real-world edge applications.”
“This collaboration represents a new era in AI hardware innovation. EMASS has recently transitioned away from MRAM technology because ReRAM is better able to support next-generation systems in IoT, automotive, and consumer electronics. By combining Weebit’s cutting-edge ReRAM with our ultra-low-power AI technology, we are setting the stage for a next-generation solution that will redefine energy efficiency for AI applications. Such integration can enhance system performance and also ensure scalability and sustainability, paving the way for smarter, more autonomous edge devices. With this synergy, we are poised to deliver unparalleled advancements in AI computing, driving meaningful impact across industries such as IoT, healthcare, automotive, and industrial automation.”
“Weebit ReRAM will provide customers using our 130nm BCD process with a very low-power, high-density and cost-effective NVM. Given Weebit’s high-quality technology and unique combination of design and process engineering, we believe the technology transfer and qualification processes will proceed quickly. Looking ahead, we’re delighted to see Weebit’s strong commitment to continued roadmap innovation.”
“Weebit ReRAM is an innovative Non-Volatile Memory technology that can provide IC companies with a differentiated embedded solution. As a Weebit design partner, IC’Alps can support the unique design needs of Weebit’s customers to help them quickly achieve working silicon and speed their time to market for SoCs integrating this advanced NVM.”
“The work Weebit and CEA-Leti are doing to make Weebit ReRAM available on GlobalFoundries’ 22FDX is a welcome development as we continue to expand the ecosystem around this platform. Embedded NVM is a key element of our customers’ designs, but since embedded flash is difficult to scale below 28nm, many customers are looking to NVM solutions such as embedded ReRAM.”
“Through our ongoing work with Weebit, we are getting closer to the realization of truly brain-like platforms. The AI system using our algorithms and Weebit ReRAM combines the accuracy of deep learning with the flexibility of the human brain, so it can learn new things without forgetting trained tasks of previously acquired information. Through our close collaboration with Weebit, we have thoroughly tested their ReRAM samples and can assure their performance, high accuracy and low-power profile.”