The objective of this research is to develop cross-layer co-designed computing systems that mitigate the memory wall in modern data-intensive workloads by reducing data movement and improving energy efficiency through coordinated innovations across algorithms, architecture, circuits, and emerging memory technologies. As workloads such as deep neural networks, large-scale vector search, and scientific data analytics continue to scale, system performance is increasingly limited by memory bandwidth rather than compute capability, creating significant latency and energy bottlenecks. To address this challenge, this research proposes several complementary computing frameworks. First, Axon, a novel systolic array architecture, improves in-array data orchestration and enables efficient hardware support for convolution operations, reducing runtime and memory traffic with minimal hardware overhead. Second, HyDra, a hyperdimensional computing accelerator based on spin–orbit torque content-addressable memory (SOT-CAM), performs key operations such as binding, permutation, and similarity search directly within memory arrays to minimize data movement. Third, HERP, a system-level platform, demonstrates the practical benefits of these approaches by accelerating large-scale database search and clustering for mass-spectrometry-based proteomics. Future work will extend Axon with multi-GEMM streaming using a “read-as-ready” scheduling method and evaluate it against state-of-the-art accelerators toward chip tape-out. The research will also explore scalable in-memory approximate nearest neighbor search (ANNS) using SOT-CAM and investigate a 3-bit weighted XOR-CAM unit to further improve the accuracy and efficiency of in-memory similarity search engines.