Professor JP Wang of the University of Minnesota presented at the Vanderbilt Seminar Series.
The resource examines emerging in‑memory computing architectures that integrate data storage and processing within the same physical element, a paradigm motivated by the escalating power and bandwidth demands of AI workloads. By focusing on radiation‑hard applications, the discussion connects fundamental device physics to the stringent reliability requirements of space‑borne and defense systems. The narrative situates this approach within the broader evolution of computing from the transistor era to modern GPUs, TPUs, and edge accelerators, highlighting the need for fundamentally more efficient solutions.
Technical coverage includes a detailed analysis of magnetic random‑access memory (MRAM) technologies, particularly perpendicular magneto‑tunnel junctions and spin‑orbit‑torque (SOT) MRAM, and their suitability for high‑density, radiation‑tolerant operation. The presenter contrasts the energy per bit of logic (≈10⁻¹⁸ J), high‑bandwidth memory (≈10⁻¹⁴ J), and data transfer (≈10⁻¹¹ J), illustrating why conventional GPU‑centric designs become untenable at scale. The architecture leverages true random‑access memory cells that retain programmability while performing compute primitives directly, eliminating costly data movement. Experimental results from collaborations with JPL, NIST, and industry partners demonstrate ultra‑fast switching under radiation exposure and the feasibility of integrating these devices into standard CMOS processes.
With this knowledge, learners can articulate how memory cells execute computation, evaluate the comparative energy efficiency of traditional GPU‑based systems versus in‑memory computing platforms, and recognize key emerging devices such as MRAM that enable radiation‑hard operation. This understanding equips them to assess design trade‑offs for edge AI, space electronics, and high‑performance computing, and to guide future research or development of architectures that fuse storage and processing for superior power efficiency.