Memristors and HPE Machine Focused Computer Idea Seems Almost Dead

There was a lot of hype and promise around memristors and HPE bet huge amounts of research around creating a new computer paradigm around memristors and memory driven computing. The last major flurry of information and presentations from HPE was in 2018.

From 2014-2018, HPE launched memory driven computing that was supposed to leverage memristor capabilities to create a revolution in computing. In April 2018, Travelport began the first commercial Memory-Driven Computing engagement, installing an HPE Superdome Flex in-memory computing system and leveraging HPEs expertise to rearchitect key Travelport algorithms using Memory-Driven Computing programming techniques.

In 2020, HPE Superdome Flex offers modular, highly flexible, and reliable platform that scales from 4 to 32 processors as a single system and delivers 768 GB to 48 TB of shared memory. HPE Memory driven computing looks like regular servers with a larger memory configuration.

There seems to have been virtually no talk about the Machine project from HPE. In 2015, the pure memristor was removed from the HP roadmap.

Knowm is a small startup that is still working on memristors. There seems to be academic research for few million to a few tens of millions on memristors. It seems the current memristors are not good enough to actually create the neuromorphic computing revolution or the nanoscale integration of memory and computing.

There is academic work on trying to realize the promise of computers that are thousands of times more energy efficient using memristors.

Let me know in the comments if I missed something. Memristors seem to be dead or we are in a memristor-winter until needed research breakthroughs change the situation.

There was a 2019 German report of the future of memristor computing.

They claim several NVM (non-volatile memory) technologies belong to Chua’s memristor class:
• PCM (Phase Change Memory), which switches crystalline material, e.g. chalcogenide glass, between amorphous and crystalline states by heat produced by the passage of an electric current through a heating element,
• ReRAM (Resistive RAM) with the three sub-classes
– CBRAM (Conductive Bridge RAM), which generates low resistance filament structures between two metal electrodes,
– OxRAM (Metal Oxide Resistive RAM), in which the thickness ratio between the resistance switching layer and the base layer in a bi-layer bi-layer oxide structure with nearly stoichiometric, oxide layer (resistance switching layer) with higher resistivity and a metal-rich layer with lower resistivity (base layer) is changed by redistribution of oxygen vacancies, and
– DioxRAM (Diode Metal Oxide Resistive RAM), in which oxygen vacancies are redistributed and trapped close to one of the two metal electrodes and lower the barrier height of corresponding metal electrode,
• MRAM (Magnetoresistive RAM)storing data by magnetic tunnel junctions (MTJ), which is a component consisting of two ferromagnets separated by a thin insulator,
• STT-RAMs (Spin-Transfer Torque RAMs) as newer technology that uses spin-aligned (“polarized”) electrons to directly torque the domains, and
• NRAM (Nano RAM) based on Carbone-Nanotube Technique.

All commercially available memristive memories feature better characteristics than Flash, however, are much more expensive.

Written By Brian Wang,