The energy cost of computing and artificial intelligence (AI) is spiraling out of control, forecast to reach 20.9% of global energy consumption by 2030. Training a neural net to robotically solve a Rubik's Cube toy consumed 2.8 GWh , while human brains consume just ~20 W. The recent successes of large machine learning models such as OpenAI's GPT-3 and Chat-GPT are accompanied by huge carbon footprints - Chat-GPT consumed $15 million in electricity during training & generated ~552 tons of CO2 . Its ongoing energy bill is estimated at ~$3 million/month, with accompanying levels of greenhouse emissions. This unsustainable energy consumption represents both a real barrier to reaching net-zero futures and a ceiling on the power of AI computing. A big part of this problem is that we're currently trying to do brain-like computing with computers that are nothing like a brain. Today's computers use far more energy shuttling data between separate memory and processor units than actually processing, whereas neurons in the brain provide integrated memory and processing - a key driver for their radically lower energy cost. Consequently, there is a pressing need for hardware systems that function in a brain-like (neuromorphic) manner, storing and processing information natively in the same unit. In many ways, nanomagnets behave a lot like neurons in the brain. They can react to the behaviour of surrounding magnets, flipping their poles from north to south similar to how neurons send jolts of electricity. Nanomagnets can remember what they've seen in the past and change their behaviour in response to this, learning from their experiences and gradually improving at tasks like voice recognition and pattern prediction. Nanomagnets provide both memory from their ability to remember data for 1000s of years (hard drives were originally made from nanomagnets for this reason), and processing from their ability to react nonlinearly to input data at GHz speeds - oscillating in a special way known as 'magnonics'. Indeed, the maths powering modern software neural networks originate from theoretical frameworks developed by physicists in the 1970's to describe strongly-interacting magnetic networks . The early machine learning community adopted these frameworks (originally termed Hopfield networks ) and adapted & refined them into the neural networks of today. Since the early successes of machine learning, engineers have dreamt of removing the software layer of abstraction and implementing machine learning directly in physical magnetic networks. However until recently, the engineering challenges of providing efficient data input and output schemes had prevented realisation of such systems. Our team have now solved these issues to accomplish the world-first example of neuromorphic computing in nanomagnetic arrays, using the magnon dynamics of a nanomagnetic array to process information and solve a range of AI tasks including future prediction of complex biological signals. We now have a way to massively improve the power of our AI computation at no extra energy cost, by moving our nanomagnetic arrays from 2D into 3D structures, our early results and simulations show that our computing power is likely to radically increase. In this project, we will work between a group in the UK led by early-career researcher Jack Gartside and a group in the USA lead by world-expert Prof. Benjamin Jungfleisch to test our ideas & bring low-energy, low-carbon AI one step closer to reality.

In order to function, all cells in the body require a regular supply of oxygen and continuous removal of waste products. Both are provided by blood delivered through the microvasculature, which comprises vessels smaller than 0.1 mm in diameter. In order to fulfil its function, the flow of blood must be tightly regulated. A key component of this regulation are the specialist 'endothelial cells' that line all microvessels. These cells sense frictional forces arising from the flowing blood and in response release chemical substances that can increase or decrease the size of the vessels to help regulate the flow. When this regulation fails, the results can be devastating. For example, dysregulation of blood flow is one of the first stages in diabetic retinopathy, a condition that threatens the sight of 1% of the world's adult population. It is therefore important to understand the details of how blood flows in microvessels. A major factor that influences microvascular blood flow is the mechanical properties of red blood cells (RBCs). RBCs are highly deformable, which allows them to deform while flowing in larger vessels and even fit through capillaries much smaller than their diameter. RBCs also have a propensity to stick together, in a process called aggregation that is dependent on local flow characteristics. As a result of these RBC behaviours, the flow of blood in microvessels is complex and poorly understood. This is particularly important, because in numerous microvascular diseases, including diabetes, the RBCs become less deformable and aggregate more than in healthy individuals. These changes have been shown to correlate with disease progression, but it has not yet been established exactly how changes to blood properties affect microvascular function. We hypothesise that the changes in RBC properties alter blood flow and hence the frictional forces experienced by the endothelial cells, which in turn leads to dysregulation of flow and ultimately damage to the microvasculature. In this project, we will use state-of-the-art experimental technology to directly evaluate how changes to RBC properties affect microscale blood flow. A key challenge is the complicated branching patterns of the microvessel network. These networks consist of vessels of different sizes, structure and functions, throughout which both RBC flow and concentration change significantly. In order to improve our knowledge of how blood flows in microvessels, we need to be able to measure both the velocity of the RBCs and their local concentration in a given blood vessel or section of a microvascular network. We will achieve this using recently developed optical techniques, combining measurements of light passing through a blood sample with fluorescence measurements of microparticles added to the plasma. Acquiring both of these parameters allows calculation of the frictional forces on the vessel wall, which will be compared to results generated with numerical models. It is not currently possible to make these measurements in humans or living animals, hence we will build realistic models of microvessels using a new technique where laser energy is used to degrade a hydrogel, leaving behind a vessel structure that can be precisely controlled. We will flow blood from healthy volunteers through these models and measure the flow and wall friction under various conditions. We will then chemically treat the blood samples to mimic changes that occur in diabetes and measure the corresponding changes in flow. In addition to providing new insight into blood flow, the evidence generated in this study will reveal how changes to blood mechanical properties might affect diseases such as diabetes. In the long term, this insight is expected to lead to new approaches for diagnosing and treating microvascular diseases.

Context: Fuel cells - why are they important? Fuel cells are devices that are able to produce electricity for transport, industrial and residential applications directly from electrochemical reactions. Among fuel cells, proton-exchange membrane fuel cells (PEMFCs) are one of the most promising, since hydrogen is used to produce electricity that can be used to power an electric car, or a home. Fuel cells produce electricity very efficiently, and the use of hydrogen produces fewer greenhouse gases than does burning fossil fuels. This also helps to preserve energy resources, as well as to produce water as the only byproduct of the electrochemical reactions, which is a clear benefit for the environment. However, hydrogen is not found freely in nature and must be extracted from other sources. In addition, hydrogen is a gas and presents several issues in terms of safety (handling, transport and storage). Another important drawback of PEMFCs is the use of costly noble metals as catalysts, such as Pt and Pd. All these factors are an obstacle for full exploitation and implementation of PEMFCs. What do novel hydroxide exchange membrane fuel cells (HEMFCs) have to offer? The most significant advantage of HEMFCs is that under alkaline conditions, electrode reaction kinetics are much more facile, allowing the use of inexpensive, non-noble metal catalysts, such as NiO and CoO. Another key advantage is that while in acidic conditions as in PEMFCs corrosion is an important issue, instead in alkaline media as in HEMFCs, corrosion is substantially reduced. More importantly, alkaline media are favourable for the use of methanol or ethanol as a fuel. Methanol is very attracting in fuel cells because he has higher volumetric energy density compared to hydrogen and its storage and transportation is less problematic than hydrogen. Also, methanol crossover is reduced in HEMFCs compared to PEMFCs, due to the opposite direction of ion transport in the membrane, from the cathode to the anode. These characteristics make the HEMFC technology economically viable and competitive within internal combustion engines. The polymer utilised herein (TPQPOH) is very competitive in terms of costs (e.g. ~£1/m2 vs. ~£500/m2 for Nafion) and durable in an alkaline environment and additional advantages could be obtained when this polymer is used as a composite material along with carbon nanomaterials. Impact The biggest challenge in developing alkaline fuel cells is the anion exchange membrane. Typically, anion exchange membranes are composed of a polymer backbone with tethered cation exchange groups, in order to facilitate the transport of hydroxide ions. The role of the anionic exchange membranes is very similar to the role of Nafion membrane in PEMFCs, where a sulfonic (anion) group is covalently attached to the polymer backbone and protons travel from the anode to the cathode through the membrane. However, in HEMFCs , hydroxide ions travel through the membranes instead of protons, and the challenge is to fabricate membranes with high hydroxide conductivity, good mechanical stability and resistance to chemical deterioration at high temperatures. Another challenge is obtaining values of hydroxide conductivity comparable to proton conductivity observed in PEMFCs. The lack of effective hydroxide exchange membranes is one of the major obstacles to the development of HEMFCs. Long-term development could generate impact through the development of novel composite materials including TPQPOH/carbon nanomaterial (single- and multi-walled carbon nanotubes and graphene) derivatives. More importantly, the use of doped graphene derivatives as catalyst will enable the development of metal-free fuel cells without the use of precious metal catalysts with an obvious beneficial impact in terms of costs. By switching from internal combustion engines to fuel cells, it is very clear how significant developments in fuel cells could have a dramatic positive impact to our society.

A key objective of computational materials science is to relate molecular structure to material properties. In many areas of materials science this approach now provides a standard tool for the controlled design of novel materials. However, the prediction of the structural and thermodynamic properties of protein-based materials is much more challenging because in this case all-atom models are too expensive to be viable for the prediction of phase behavior and kinetics. The aim of the present work is to use an integration of experiments (US partners Fraden and Lenhoff) and simulation (UK partner Frenkel and US partner Lenhoff) to develop and validate an approach that will make it possible to predict the phase behaviour and crystallisation pathways of proteins in solution. This proposed research is motivated by the observation that protein-based materials play a key role in science and technology. To start with a simple observation: our bodies contain high concentrations of proteins and these proteins have evolved to perform either "biochemical" or "structural" tasks (the division is not always sharp). In science, the preparation of high-quality protein crystals is a crucial step in the elucidation of 3D protein structures by X-ray or neutron diffraction. In addition, the thermodynamic and structural properties of protein-based products are of key importance for the shelf life and bio-avalability of many pharmaceuticals and food products. It is therefore clearly desirable to be able to predict the phase behaviour and structural properties of protein-based materials on the basis of microscopic information. At present, our ability to make such predictions is limited by the absence of reliable predictive tools. The objective of the proposed research is to pool the expertise of three world-leading groups in the area of protein crystallization and gelation, to develop modelling techniques that will allow us to predict structure, crystal nucleation and phase stability of protein systems. The project will combine the expertise of two US groups and one UK group. The US groups comprise an expert on experimental studies of protein phase behaviour (Lenhoff, Delaware) and a leader in the field of microfluidics-based protein crystallization (Fraden, Brandeis). The UK group (Frenkel, Cambridge) has a strong track record in the numerical modelling of protein phase behaviour and crystal nucleation. The key objective of the proposal is develop a systematic procedure that allows us to construct simplified, but physically meaningful, molecular models of proteins for computer simulations. These models should be sufficiently refined that they will allow us to predict/elucidate experimental studies of the equilibrium phase diagrams and phase separation kinetics of protein solutions. Clearly such a model will need to be validated extensively. Our project will therefore be based on a tight coupling between modelling and experimental validation. Measurements of nucleation rates will be interpreted in the context of the classical nucleation model and measurements of the growth of individual precritical nuclei will be used to test the assumptions of that model. The effect of kinetically arrested states, such as non-equilibrium gels and precipitates, on crystallization will be studied in both experiment and simulation. The broader impact of the proposed research will be that it will enable a better control of the properties of protein-based materials. In particular, it should allow us to improve the rate of protein crystallization for structural biology and to control protein aggregation for pharmaceutical applications.

Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.