Will Lithium-based Memristors Spur an Analog Computing Renaissance?

October 04, 2021 by Adrian Gibbons

In an AAC interview with MIT's faculty Jennifer Rupp and Ericsson's Dr. Saeed Bastani, we learned how a new field, "lithionics," could be the key to advancing neuromorphic computing.

As it turns out, the road to neuromorphic computing will hinge heavily on materials science. Lithionics, a new field emerging from MIT, investigates how lithium oxides may be such a material. Additionally, researchers from both MIT and Ericsson are exploring how memristors, or memory resistors, based on lithium oxide films may be the key to creating computer chips that mimic the human brain.


Integrated memristive neural network

Example of an integrated memristive neural network from a 2018 Nature Electronics review of memristors. This optical micrograph shows an 8 × 8 memristive synapse crossbar. Image used courtesy of Nature Electronics

In a follow-up interview to our initial discussion of this "frontier research," All About Circuits sat down with two principal researchers to learn more about lithionics and associated memristor technology.  

These two leading researchers are Dr. Jennifer L.M. Rupp, an associate professor of materials science and engineering and electrical engineering and computer sciences at MIT, and Dr. Saeed Bastani, a researcher at Ericsson working on the applications of lithionics, including neuromorphic computing.

The two experts shared their insights on lithium as computing technology, a lithium processor's potential for system-level integration, and the promise of memristor technology in developing artificial intelligence systems.


A Window Into Lithionics Research

While silicon is the go-to material for neuromorphic computer chips, researchers are searching for a material that uses exponentially less energy. Lithium oxides are one promising alternative for future memristors-on-chip. 

“Our new team of Professors Sze, Bazant and del’ Alamo, and myself and Dr. Balaish at MIT, is making lithium-based materials in the range of hundreds of nanometers in thickness on chips,” Dr. Rupp says. “After jumping recently to the software side for machine learning analysis of high throughput material synthesis, we don’t have enough computational power in hardware." The joke, she explains, is that she needs her own future chips to analyze the vast amount of materials research she is currently working on. 


Dr. Rupp

Dr. Jennifer L.M. Rupp. Image used courtesy of Dr. Rupp


In data published in Nature Review Materials and Advanced Materials, MIT Professors Rupp and Bazant and their teams evaluated two different lithium-based films against several criteria, including:

  • Conductance
  • I–V cycling response
  • Resistance retention time of the films under low bias

The results indicated that one of the two films, Li7Ti5O12, would be appropriate for deep neural networks (DNNs) while Li4Ti5O12 was more suited to spiking neural networks (SNNs) applications.

Dr. Rupp and her team intend to explore just how challenging it might be to integrate lithium-based computing hardware with existing silicon-based technology. "There’s this existing technology [silicon], that is very far developed and you don’t want to change that too much," she comments.

Dr. Rupp also mentions another potential spin-out from lithionics research—the possibility of “micro-batteries” on silicon down to the microchip level.


Lithium Memristors: A Critical Step to Analog Computers?

One of the focal points of the MIT and Ericsson lithionics research is the development of lithium-based memristors. Memristors are a technology that joins processor operations and memory into the same architecture. MIT has previously conducted research on memristors, which included integrating thousands of these artificial brain synapses on a single chip. Interestingly, this technology, which has gained research interest in recent years, could point to an analog computing renaissance.

Memristors are said to mitigate two primary drawbacks of digital systems: 1) the movement of data to and from processor to memory and 2) the limitation of a transistor's bi-stable logic. 


Example of a memristive linear multilevel synaptic device

Example of a memristive linear multilevel synaptic device. Image used courtesy of the Journal of Applied Physics


Dr. Bastani explains that memristors may become a critical key to advancing neuromorphic computing. Using the example of solving a linear equation, Bastani says, “Depending on the application characteristics, certain performance levels need to be offered by the device. For instance, a certain number of reliable conductance states needs to be supported.”

On the discussion of precision and multi-level conductance, Dr. Bastani notes that elaborate mechanisms can be devised to compensate for the precision limitation of a memristive device. "We don’t know yet what precision level will be offered by the lithionic device. We at Ericsson are doing research to find possible ways to compensate for potential limitations," he explains. "For example, algorithm-hardware co-design is a promising methodology to tackle some limitations, including those related to device precision."


MIT and Ericcson Divide and Conquer Lithium-based Memristors

How exactly will this lithionics research unfold between MIT and Ericsson? According to a recent press release, four teams at MIT will divide and conquer the task.

Professor Martin Bazant will head one team's efforts to find an optimal lithium composition for computing using computational modeling. A second team, led by Dr. Rupp and her colleague Moran Balaish, will create the new materials and integrate them into a memristor prototype. 

Meanwhile, a third team spearheaded by Professor Jesus del Alamo will explore on-chip integration technology. The fourth team led by Professors Vivienne Sze and Joel Emer will investigate computer architectures that may benefit from the lithium-based memristors. 

Finally, a team at Ericsson will double down on the applications of such devices, evaluating specific algorithms and hardware architectures comprising the proposed lithionic chips.


It’s an Analog World After All

Analog computing is not the first technology to come back into fashion. Consider, for instance, that the first cars were electric

Memristor-based computing appears to be the next stage in the development of analog computers. However, instead of being room-sized, these computers will be scaled down to the size of digital computers thanks, in part, to advances in material science—like lithionics.


Crossbar memristors

Memristor crossbar arrays allow mapping and execution of convolutional layers. Image used courtesy of Applied Physics Review


The aim of lithionics research is to develop specialized hardware to ramp up computing speeds while lowering power demands. To determine the viability of lithium-based materials for memristor applications, researchers must also find a balance of tuning symmetry and retention characteristics in lithium oxide materials. 

Nearly 75 years after the invention of the transistor and the start of the digital era, computer engineering appears to be coming full circle. It turns out it may be an analog world after all.