The MemOD project builds computer memory that works more like a human brain, using custom-designed molecules stacked into ultra-thin films. Modern computers waste enormous energy shuttling data between separate processing and memory units—a bottleneck that forces even a champion Go-playing AI to gulp nearly a megawatt of power while its human opponent runs on 20 watts and a cup of tea. The solution is to perform calculations directly inside memory using memristors, devices that mimic the brain’s synapses by storing and processing information in one place. But today’s oxide-based memristors are unreliable: they form random conductive filaments that degrade over time, limiting the global market to just $200 million annually. MemOD sidesteps this randomness by building memristors from ordered organometallic molecules, assembled layer by layer with atomic precision. The team will exploit quantum interference effects within these molecules to boost performance, then wire them into devices using chemical anchor groups and a graphene top contact. If successful, this fundamental science could unlock energy-efficient AI hardware, slashing the power demands of data centres, autonomous systems, and edge devices—without waiting for a distant breakthrough.
View original technical description
Imagine a world where machines learn not through energy-hungry programmed algorithms, but by forging connections and adapting like the human brain. A world where artificial intelligence systems can seamlessly integrate vast and complex data streams, make intuitive decisions, and continually evolve their understanding of the world around them. In the MemOD project, we aim to establish a new class of highly ordered, ultrathin and efficient materials which will help make this world a reality. The challenge is that modern computers require data to be transferred between a central processing unit (CPU) and memory, in order to perform a computation. This data transfer is known as the von Neumann bottleneck, and not only limits the speed of computation, but is also highly energy-inefficient. For a direct comparison between human and machine, we can consider the ancient strategy game, Go. AlphaGo, a brain-inspired computer owned by Google, eventually defeated the Korean Go world champion a few years ago. However, while AlphaGo consumes almost a megawatt of power, the Korean champion needed a mere 20 watts and the energy to make a cup of tea. The solution is to perform computation in memory, thereby reducing the machine's energy cost. This is accomplished by utilising memristors, which are low-power devices, able to simulate the synapses in our brain, bypassing the inefficient von Neumann bottleneck. Memristors are electrical elements whose resistance can be programmed. They store this state even if the device loses power. A stable, tuneable, efficient memristor is the holy grail of AI deployment. Although tantalisingly close, such a memristor has yet to be realised, because in traditional oxide-based inorganic devices, memristive states are due to the formation of conductive filaments between the device electrodes and unfortunately, this process is random, resulting in device variability and signal degradation over time. These deficiencies are why the global memristor market is currently only $200 million per annum. However, this market is predicted to reach $2.5 billion by 2028, on the assumption that these problems can be solved, for example, through the development of molecular memristor technology. The MemOD project will utilise ordered films of organometallic molecules as the building blocks of a new class of memristors which are not limited by this random mechanism, to overcome these deficiencies. We will design and synthesise novel organometallic molecules that we will build up into highly ordered self-assembled thin multilayers using sequential deposition, to enable precise control over composition and properties, thereby decreasing the random nature of memristive switching. Furthermore, as demonstrated by the applicants, we will utilise quantum interference effects taking place within organometallic molecules to increasing the on/off ratio and other figures of merit. The resulting organometallic memristors will be wired into devices via chemical anchor groups, which offer accurate control of the contact to device-compatible electrodes fabricated from CMOS (complementary metal-oxide semiconductor) materials, resulting in lower variability, and will be contacted by a graphene layer on top. The applicants have the ideal combination of world-leading expertise spanning molecular modelling, design, synthesis, characterisation and device integration. They have a proven track record of innovation and successful collaboration (46 joint papers, with >2,300 citation as of 09/23) and are uniquely placed to deliver this ambitious project.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know