Neural network (NN) inference is an essential part of modern systems and is found at the heart of numerous applications ranging from image recognition to natural language processing. In situ NN accelerators can efficiently perform NN inference using resistive crossbars, which makes them a promising solution to the data movement challenges faced by conventional architectures. Although such accelerators demonstrate significant potential for dense NNs, they often do not benefit from sparse NNs, which contain relatively few non-zero weights. Processing sparse NNs on in situ accelerators results in wasted energy to charge the entire crossbar where most elements are zeros. To address this limitation, this letter proposes Granular Matrix Reordering (GMR): a preprocessing technique that enables an energy-efficient computation of sparse NNs on in situ accelerators. GMR reorders the rows and columns of sparse weight matrices to maximize the crossbars' utilization and minimize the total number of crossbars needed to be charged. The reordering process does not rely on sparsity patterns and incurs no accuracy loss. Overall, GMR achieves an average of 28 percent and up to 34 percent reduction in energy consumption over seven pruned NNs across four different pruning methods and network architectures.
Atomic precision advanced manufacturing (APAM) offers creation of donor devices in an atomically thin layer doped beyond the solid solubility limit, enabling unique device physics. This presents an opportunity to use APAM as a pathfinding platform to investigate digital electronics at the atomic limit. Scaling to smaller transistors is increasingly difficult and expensive, necessitating the investigation of alternative fabrication paths that extend to the atomic scale. APAM donor devices can be created using a scanning tunneling microscope (STM). However, these devices are not currently compatible with industry standard fabrication processes. There exists a tradeoff between low thermal budget (LT) processes to limit dopant diffusion and high thermal budget (HT) processes to grow defect-free layers of epitaxial Si and gate oxide. To this end, we have developed an LT epitaxial Si cap and LT deposited Al2O3 gate oxide integrated with an atomically precise single-electron transistor (SET) that we use as an electrometer to characterize the quality of the gate stack. The surface-gated SET exhibits the expected Coulomb blockade behavior. However, the gate’s leverage over the SET is limited by defects in the layers above the SET, including interfaces between the Si and oxide, and structural and chemical defects in the Si cap. We propose a more sophisticated gate stack and process flow that is predicted to improve performance in future atomic precision devices.