A New Nonvolatile Switch for the Age of AI
Please join us at UCCS For Our 2nd Electronic Devices/Circuits & Systems Meeting of 2024. Hear information on the latest storage developments from the UCCS Microelectronics Labs and enjoy lunch provided by the Pikes Peak Section.
A New Nonvolatile Switch for the Age of AI
Prof. Carlos Araujo, Distinguished Professor, IEEE Fellow
University of Colorado-Colorado Springs ECE Department
Memory and artificial synapses are the two areas that require major innovation for the next generation of AI engines. Today’s memory of choice for the highest performance AI systems in High Bandwidth Memory (HBM) providing incredible bandwidth in a small space. A Micron HBM3E has an energy efficiency quoted at 1E-12 J per switching operation. While good, there is much concern about the energy consumption of AI data centers, some which may require a nuclear power plant to supply their energy.
In this lecture, I describe a non-volatile switch that has been made at UCCS and IMEC using the Mott Quantum Switch (MQS) at 45 nm design rules with energy performance of 1E-18 J. It has the potential to scale to the most advanced feature size using nanosheets made using common gate oxide Hafnium Oxide (HfO2) with doping with Carbon. These fast switches have excellent nonvolatile properties from 2 degrees Kelvin to 200 degrees C with less than 2 ns switching times and require only 3-4 mask steps. This makes them ideal for low power nonvolatile storage in future AI systems. For synapses used in neural nets, these switches have programmable levels of resistivity in a deterministic way for programmable neuron weights.
Date and Time
Location
Hosts
Registration
- Date: 31 Oct 2024
- Time: 12:00 PM to 01:30 PM
- All times are (UTC-06:00) Mountain Time (US & Canada)
-
Add Event to Calendar
- 1420 Austin Bluffs Parkway
- Colorado Springs, Colorado
- United States 80918
- Building: Engineering Building, 2nd Floor
- Room Number: Tom Saponas Lounge
- Starts 27 October 2024 04:00 PM
- Ends 31 October 2024 10:00 AM
- All times are (UTC-06:00) Mountain Time (US & Canada)
- No Admission Charge
Speakers
Carlos Araujo
A New Nonvolatile Switch for the Age of AI
A New Nonvolatile Switch for the Age of AI
Prof. Carlos Araujo, Distinguished Professor, IEEE Fellow
University of Colorado-Colorado Springs ECE Department
Memory and artificial synapses are the two areas that require major innovation for the next generation of AI engines. Today’s memory of choice for the highest performance AI systems in High Bandwidth Memory (HBM) providing incredible bandwidth in a small space. A Micron HBM3E has an energy efficiency quoted at 1E-12 J per switching operation. While good, there is much concern about the energy consumption of AI data centers, some which may require a nuclear power plant to supply their energy.
In this lecture, I describe a non-volatile switch that has been made at UCCS and IMEC using the Mott Quantum Switch (MQS) at 45 nm design rules with energy performance of 1E-18 J. It has the potential to scale to the most advanced features using nanosheets made using common gate oxide Hafnium Oxide (HfO2) with doping with Carbon. These fast switches have excellent nonvolatile properties from 2 degrees Kelvin to 200 degrees C with less than 2 ns switching times and require only 3-4 mask steps. This makes them ideal for low power nonvolatile storage in future AI systems. For synapses used in neural nets, these switches have programmable levels of resistivity in a deterministic way for programmable neuron weights.
Biography:
Carlos A. Paz de Araujo is the Associate Dean of Research Development Electrical and Computer Engineering at UCCS. He was a founder of Ramtron and Symetrix Corporations which pioneered the ferroelectric RAM. He won the IEEE Daniel E. Noble Award in 2006 for contributions to emerging technologies.
