"Computational Memory" Computing | Tensor Flow + Hyperdimensional Computing

"Computational Memory" Computing | Tensor Flow + Hyperdimensional Computing

This session introduces the attendee to the concept of "Computational Memory". The idea is to couple, fast memory based algorithms to data contained within the physical semiconductor memory device itself. That the compute capability lies within the "In-Memory" paradigm is seen as the most advantageous architecture to segue from von Neumann into "Pattern Computing". 

Coupling of this memory based architecture with new, emerging semiconductor memory technologies such as 3DXPoint (Optane) and Vertical Resistive RAM (VRRAM) is discussed as are the programs that are currently in pursuit of the goal of providing a scalable, In-Memory Tensor Flow and Hyperdimensional Computing (HDC) Use Adaptation for the Cloud and Enterprise Computing environments. The idea of this computational facility being further applied to "Computational Storage" includes an NVMe based Optane Memory Drive Fabric discussion.

The final discovery in the stack covers the ability of HDC to accelerate the search, sort, conjugate and clustering functions of data into Affinity Propagation Stacks useful in Predictive Analytics, Natural Language Processing, Vertical Search, and Cognitive Synthetic Intelligence application use cases. 




Regency Ballroom A


Principal Analyst, WebFeet Research Inc.
WebFeet Research Inc.
Gil Russell merged Semiscape with WebFeet Research Inc. in August 2016. He continues his research into semiconductor memory technology and their optimum utilization in hardware/algorithmic frameworks for both machine learning and general intelligence classes of computing. This includes these sub-derivative elements;

• Near Data Processing Acceleration
• Predictive Analytics
• Neuromorphic Architectures
• In-Memory Computing Cognitive Extensions

Current research is focused on investigating the adaption of In-Memory Compute machine server architecture to High-Dimensional Computing within a standard rack model space – leading to the development of a standard hyperscale cognitive base element for Cloud Computing.
WebFeet Research
Alan Niebel is the CEO for WebFeet Research that has been recognized as the premier market research firm that focuses exclusively on the memory and storage markets. WFR’s charter qualifies and quantifies Flash and non-volatile memory and bridges their transition to next generation use and cognitive computing. Alan’s over 35 year career in electronics includes over 20 years as an analyst where he writes, speaks and consults with every major semiconductor manufacturer. His current projects analyze emerging market solutions for In-Memory Computing inclusive of first generation persistent memory platforms. Subjects covered include the application of intelligence to direct attached Flash Storage, Memory Channel NVDIMM and DRAM/NVRAM/XPoint technologies.

Slides & Recordings

   Download Slides