This inventor holds 1 USPTO granted patent and 1 published patent application. Top assignee: University of Michigan. Active years: 2025.
Company Filing History:
Years Active: 2025
Title: Innovations of Nishil Talati in Hybrid Memory Systems
Introduction
Nishil Talati is an accomplished inventor based in Ann Arbor, MI (US). He has made significant contributions to the field of memory systems, particularly in the development of hybrid memory solutions that enhance the performance of recommendation models. His innovative approach addresses the limitations of traditional memory systems, paving the way for more efficient computing.
Latest Patents
Nishil Talati holds a patent for a "DRAM-RRAM hybrid memory system for recommendation models." This invention tackles the performance limitations of modern recommendation models, which often struggle due to the memory bandwidth-hungry embedding layer reductions. The hybrid memory system combines DRAM and RRAM with processing-in-memory (PIM) capabilities. It presents a comprehensive optimization approach that includes access-pattern aware mapping, compute complexity reduction, and selective PIM reduction to mitigate computation latency. An evaluation of this system demonstrates significant improvements, achieving performance, energy, and energy-delay product (EDP) enhancements of 2.6×, 1.7×, and 4.4×, respectively, compared to a CPU baseline.
Career Highlights
Nishil Talati is affiliated with the University of Michigan, where he continues to push the boundaries of memory technology. His work is characterized by a strong focus on optimizing memory systems for modern computational needs. His innovative contributions have garnered attention in both academic and industry circles.
Collaborations
Nishil has collaborated with notable colleagues, including Heewoo Kim and Haojie Ye. Their combined expertise has further advanced the research and development of hybrid memory systems.
Conclusion
Nishil Talati's work in hybrid memory systems represents a significant advancement in the field of memory technology. His innovative patent addresses critical performance issues in recommendation models, showcasing the potential for improved computational efficiency. His contributions are paving the way for future innovations in memory systems.