Spatial accelerators, composed of arrays of compute-memory integrated units, offer an attractive platform for deploying inference workloads with low latency and low energy consumption. However, fully exploiting their architectural advantages typically requires careful, expert-driven mapping of computational graphs to distributed processing elements. In this work, we automate this process by framing the mapping challenge as a black-box optimization problem. We introduce the first evolutionary, hardware-in-the-loop mapping framework for neuromorphic accelerators, enabling users without deep hardware knowledge to deploy workloads more efficiently. On Intel's Loihi 2, our method achieves up to 35% reduction in total latency compared to default heuristics on two sparse multilayer perceptron networks. We further demonstrate the scalability of our approach to multi-chip systems and observe an up to 40% gain in energy efficiency, without explicitly optimizing for it.

Linear recurrent neural networks enable powerful long-range sequence modeling with constant memory usage and time-per-token during inference. These architectures hold promise for streaming applications at the edge, but deployment in resource-constrained environments requires hardware-aware optimizations to minimize latency and energy consumption. Unstructured sparsity offers a compelling solution, enabling substantial reductions in compute and memory requirements–when accelerated by compatible hardware platforms. In this paper, we conduct a scaling study to investigate the Pareto front of performance and efficiency across inference compute budgets. We find that highly sparse linear RNNs consistently achieve better efficiency-performance trade-offs than dense baselines, with $2\times$ less compute and $36$% less memory at iso-accuracy. Our models achieve state-of-the-art results on a real-time streaming task for audio denoising. By quantizing our sparse models to fixed-point arithmetic and deploying them on the Intel Loihi 2 neuromorphic chip for real-time processing, we translate model compression into tangible gains of $42\times$ lower latency and $149\times$ lower energy consumption compared to a dense model on an edge GPU. Our findings showcase the transformative potential of unstructured sparsity, paving the way for highly efficient recurrent neural networks in real-world, resource-constrained environments.

As demand for computational resources reaches unprecedented levels, research is expanding into the use of complex material substrates for computing. In this study, we interface with a model of a hydrodynamic system, under development by a startup, as a computational reservoir and optimize its properties using an evolution in materio approach. Input data are encoded as waves applied to our shallow water reservoir, and the readout wave height is obtained at a fixed detection point. We optimized the readout times and how inputs are mapped to the wave amplitude or frequency using an evolutionary search algorithm, with the objective of maximizing the system's ability to linearly separate observations in the training data by maximizing the readout matrix determinant. Applying evolutionary methods to this reservoir system substantially improved separability on an XNOR task, in comparison to implementations with hand-selected parameters. We also applied our approach to a regression task and show that our approach improves out-of-sample accuracy. Results from this study will inform how we interface with the physical reservoir in future work, and we will use these methods to continue to optimize other aspects of the physical implementation of this system as a computational reservoir.1