Digital University Kerala unveils breakthrough in analogue computing

Researchers demonstrate fully functional memristor-based crossbar arrays capable of performing analogue matrix multiplications

In a major leap for analogue computing, the AI Chips Centre at the Kerala University of Digital Sciences, Innovation and Technology (Digital University Kerala) has demonstrated fully functional memristor-based crossbar arrays capable of performing analogue matrix multiplications.

The milestone represents an advancement in scalable, energy-efficient computational architectures, particularly in neuromorphic computing and AI hardware.

At the core of this breakthrough by the team led by Professor Alex James is the development of analogue crossbar arrays that use memristors — resistive memory devices that can store and process information in analogue form. These arrays excel in matrix multiplications, a critical operation in machine learning, offering high speed and efficiency.

The crossbar arrays are built around a RISC-V core, making them programmable and reconfigurable, which allows them to adapt to various configurations like 1T1R (one transistor, one resistor) and 1T4R (one transistor, four resistors). This flexibility supports a range of applications, from convolutional neural networks (CNNs) to spiking neural networks (SNNs), making them ideal for AI, neuromorphic computing, and beyond.

In tandem with these hardware advancements, the team developed Pymem, a hardware-software co-design tool. Pymem simplifies the design of analogue neural networks, enabling researchers to simulate, optimise, and validate architectures tailored for memristor-based hardware. By bridging hardware design and software development, Pymem accelerates the creation of analogue neural networks, reducing development cycles and making analogue computing more accessible.

The team has already demonstrated the practical potential of their technology, such as using crossbar arrays for super-resolution imaging, achieving over 75,000 stable conductance levels. This innovation not only enhances the reliability of analogue computations but also shows its ability to tackle complex tasks with precision. The team’s ongoing research explores configurations optimised for real-world applications, including edge computing, robotics, and health care.

The work by Prof. James and his team positions Digital University Kerala to pioneer developments in energy-efficient hardware. Their innovations — combining scalable analogue architectures with tools like Pymem — present a compelling alternative to traditional digital systems.