Tag Archives: Harish Bhaskaran

Adaptive neural connectivity with an event-based architecture using photonic processors

Leave a reply

On first glance it looked like a set of matches. If there were more dimension, this could also have been a set pencils but no,

Caption: The chip contains almost 8,400 functioning artificial neurons from waveguide-coupled phase-change material. The researchers trained this neural network to distinguish between German and English texts on the basis of vowel frequency. Credit: Jonas Schütte / Pernice Group Courtesy: University of Münster

An October 23, 2023 news item on Nanowerk introduces research into a new approach to optical neural networks

A team of researchers headed by physicists Prof. Wolfram Pernice and Prof. Martin Salinga and computer specialist Prof. Benjamin Risse, all from the University of Münster, has developed a so-called event-based architecture, using photonic processors. In a similar way to the brain, this makes possible the continuous adaptation of the connections within the neural network.

Key Takeaways

Researchers have created a new computing architecture that mimics biological neural networks, using photonic processors for data transportation and processing.

The new system enables continuous adaptation of connections within the neural network, crucial for learning processes. This is known as both synaptic and structural plasticity.

Unlike traditional studies, the connections or synapses in this photonic neural network are not hardware-based but are coded based on optical pulse properties, allowing for a single chip to hold several thousand neurons.

Light-based processors in this system offer a much higher bandwidth and lower energy consumption compared to traditional electronic processors.

The researchers successfully tested the system using an evolutionary algorithm to differentiate between German and English texts based on vowel count, highlighting its potential for rapid and energy-efficient AI applications.

The Research

Modern computer models – for example for complex, potent AI applications – push traditional digital computer processes to their limits.

…

The person who edited the original press release, which is included in the news item in the above, is not credited.

Here’s the unedited original October 23, 2023 University of Münster press release (also on EurekAlert)

Modern computer models – for example for complex, potent AI applications – push traditional digital computer processes to their limits. New types of computing architecture, which emulate the working principles of biological neural networks, hold the promise of faster, more energy-efficient data processing. A team of researchers has now developed a so-called event-based architecture, using photonic processors with which data are transported and processed by means of light. In a similar way to the brain, this makes possible the continuous adaptation of the connections within the neural network. This changeable connections are the basis for learning processes. For the purposes of the study, a team working at Collaborative Research Centre 1459 (“Intelligent Matter”) – headed by physicists Prof. Wolfram Pernice and Prof. Martin Salinga and computer specialist Prof. Benjamin Risse, all from the University of Münster – joined forces with researchers from the Universities of Exeter and Oxford in the UK. The study has been published in the journal “Science Advances”.

What is needed for a neural network in machine learning are artificial neurons which are activated by external excitatory signals, and which have connections to other neurons. The connections between these artificial neurons are called synapses – just like the biological original. For their study, the team of researchers in Münster used a network consisting of almost 8,400 optical neurons made of waveguide-coupled phase-change material, and the team showed that the connection between two each of these neurons can indeed become stronger or weaker (synaptic plasticity), and that new connections can be formed, or existing ones eliminated (structural plasticity). In contrast to other similar studies, the synapses were not hardware elements but were coded as a result of the properties of the optical pulses – in other words, as a result of the respective wavelength and of the intensity of the optical pulse. This made it possible to integrate several thousand neurons on one single chip and connect them optically.

In comparison with traditional electronic processors, light-based processors offer a significantly higher bandwidth, making it possible to carry out complex computing tasks, and with lower energy consumption. This new approach consists of basic research. “Our aim is to develop an optical computing architecture which in the long term will make it possible to compute AI applications in a rapid and energy-efficient way,” says Frank Brückerhoff-Plückelmann, one of the lead authors.

Methodology: The non-volatile phase-change material can be switched between an amorphous structure and a crystalline structure with a highly ordered atomic lattice. This feature allows permanent data storage even without an energy supply. The researchers tested the performance of the neural network by using an evolutionary algorithm to train it to distinguish between German and English texts. The recognition parameter they used was the number of vowels in the text.

The researchers received financial support from the German Research Association, the European Commission and “UK Research and Innovation”.

Here’s a link to and a citation for the paper,

Event-driven adaptive optical neural network by Frank Brückerhoff-Plückelmann, Ivonne Bente, Marlon Becker, Niklas Vollmar, Nikolaos Farmakidis, Emma Lomonte, Francesco Lenzini, C. David Wright, Harish Bhaskaran, Martin Salinga, Benjamin Risse, and Wolfram H. P. Pernice. Science Advances 20 Oct 2023 Vol 9, Issue 42 DOI: 10.1126/sciadv.adi9127

This paper is open access.

Optical memristors and neuromorphic computing

Leave a reply

A June 5, 2023 news item on Nanowerk announced a paper which reviews the state-of-the-art of optical memristors, Note: Links have been removed,

AI, machine learning, and ChatGPT may be relatively new buzzwords in the public domain, but developing a computer that functions like the human brain and nervous system – both hardware and software combined – has been a decades-long challenge. Engineers at the University of Pittsburgh are today exploring how optical “memristors” may be a key to developing neuromorphic computing.

Resistors with memory, or memristors, have already demonstrated their versatility in electronics, with applications as computational circuit elements in neuromorphic computing and compact memory elements in high-density data storage. Their unique design has paved the way for in-memory computing and captured significant interest from scientists and engineers alike.

A new review article published in Nature Photonics (“Integrated Optical Memristors”), sheds light on the evolution of this technology—and the work that still needs to be done for it to reach its full potential. Led by Nathan Youngblood, assistant professor of electrical and computer engineering at the University of Pittsburgh Swanson School of Engineering, the article explores the potential of optical devices which are analogs of electronic memristors. This new class of device could play a major role in revolutionizing high-bandwidth neuromorphic computing, machine learning hardware, and artificial intelligence in the optical domain.

…

A June 2, 2023 University of Pittsburgh news release (also on EurekAlert but published June 5, 2023), which originated the news item, provides more detail,

“Researchers are truly captivated by optical memristors because of their incredible potential in high-bandwidth neuromorphic computing, machine learning hardware, and artificial intelligence,” explained Youngblood. “Imagine merging the incredible advantages of optics with local information processing. It’s like opening the door to a whole new realm of technological possibilities that were previously unimaginable.” 

The review article presents a comprehensive overview of recent progress in this emerging field of photonic integrated circuits. It explores the current state-of-the-art and highlights the potential applications of optical memristors, which combine the benefits of ultrafast, high-bandwidth optical communication with local information processing. However, scalability emerged as the most pressing issue that future research should address. 

“Scaling up in-memory or neuromorphic computing in the optical domain is a huge challenge. Having a technology that is fast, compact, and efficient makes scaling more achievable and would represent a huge step forward,” explained Youngblood. 

“One example of the limitations is that if you were to take phase change materials, which currently have the highest storage density for optical memory, and try to implement a relatively simplistic neural network on-chip, it would take a wafer the size of a laptop to fit all the memory cells needed,” he continued. “Size matters for photonics, and we need to find a way to improve the storage density, energy efficiency, and programming speed to do useful computing at useful scales.”

Using Light to Revolutionize Computing

Optical memristors can revolutionize computing and information processing across several applications. They can enable active trimming of photonic integrated circuits (PICs), allowing for on-chip optical systems to be adjusted and reprogrammed as needed without continuously consuming power. They also offer high-speed data storage and retrieval, promising to accelerate processing, reduce energy consumption, and enable parallel processing. 

Optical memristors can even be used for artificial synapses and brain-inspired architectures. Dynamic memristors with nonvolatile storage and nonlinear output replicate the long-term plasticity of synapses in the brain and pave the way for spiking integrate-and-fire computing architectures.

Research to scale up and improve optical memristor technology could unlock unprecedented possibilities for high-bandwidth neuromorphic computing, machine learning hardware, and artificial intelligence. 

“We looked at a lot of different technologies. The thing we noticed is that we’re still far away from the target of an ideal optical memristor–something that is compact, efficient, fast, and changes the optical properties in a significant manner,” Youngblood said. “We’re still searching for a material or a device that actually meets all these criteria in a single technology in order for it to drive the field forward.”

The publication of “Integrated Optical Memristors” (DOI: 10.1038/s41566-023-01217-w) was published in Nature Photonics and is coauthored by senior author Harish Bhaskaran at the University of Oxford, Wolfram Pernice at Heidelberg University, and Carlos Ríos at the University of Maryland.

Despite including that final paragraph, I’m also providing a link to and a citation for the paper,

Integrated optical memristors by Nathan Youngblood, Carlos A. Ríos Ocampo, Wolfram H. P. Pernice & Harish Bhaskaran. Nature Photonics volume 17, pages 561–572 (2023) DOI: https://doi.org/10.1038/s41566-023-01217-w Published online: 29 May 2023 Issue Date: July 2023

This paper is behind a paywall.

Guide for memristive hardware design

Leave a reply

An August 15 ,2022 news item on ScienceDaily announces a type of guide for memristive hardware design,

They are many times faster than flash memory and require significantly less energy: memristive memory cells could revolutionize the energy efficiency of neuromorphic [brainlike] computers. In these computers, which are modeled on the way the human brain works, memristive cells function like artificial synapses. Numerous groups around the world are working on the use of corresponding neuromorphic circuits — but often with a lack of understanding of how they work and with faulty models. Jülich researchers have now summarized the physical principles and models in a comprehensive review article in the renowned journal Advances in Physics.

…

An August 15, 2022 Forschungszentrum Juelich press release (also on EurekAlert), which originated the news item, describes two papers designed to help researchers better understand and design memristive hardware,

Certain tasks – such as recognizing patterns and language – are performed highly efficiently by a human brain, requiring only about one ten-thousandth of the energy of a conventional, so-called “von Neumann” computer. One of the reasons lies in the structural differences: In a von Neumann architecture, there is a clear separation between memory and processor, which requires constant moving of large amounts of data. This is time and energy consuming – the so-called von Neumann bottleneck. In the brain, the computational operation takes place directly in the data memory and the biological synapses perform the tasks of memory and processor at the same time.

In Jülich, scientists have been working for more than 15 years on special data storage devices and components that can have similar properties to the synapses in the human brain. So-called memristive memory devices, also known as memristors, are considered to be extremely fast, energy-saving and can be miniaturized very well down to the nanometer range. The functioning of memristive cells is based on a very special effect: Their electrical resistance is not constant, but can be changed and reset again by applying an external voltage, theoretically continuously. The change in resistance is controlled by the movement of oxygen ions. If these move out of the semiconducting metal oxide layer, the material becomes more conductive and the electrical resistance drops. This change in resistance can be used to store information.

The processes that can occur in cells are very complex and vary depending on the material system. Three researchers from the Jülich Peter Grünberg Institute – Prof. Regina Dittmann, Dr. Stephan Menzel, and Prof. Rainer Waser – have therefore compiled their research results in a detailed review article, “Nanoionic memristive phenomena in metal oxides: the valence change mechanism”. They explain in detail the various physical and chemical effects in memristors and shed light on the influence of these effects on the switching properties of memristive cells and their reliability.

“If you look at current research activities in the field of neuromorphic memristor circuits, they are often based on empirical approaches to material optimization,” said Rainer Waser, director at the Peter Grünberg Institute. “Our goal with our review article is to give researchers something to work with in order to enable insight-driven material optimization.” The team of authors worked on the approximately 200-page article for ten years and naturally had to keep incorporating advances in knowledge.

“The analogous functioning of memristive cells required for their use as artificial synapses is not the normal case. Usually, there are sudden jumps in resistance, generated by the mutual amplification of ionic motion and Joule heat,” explains Regina Dittmann of the Peter Grünberg Institute. “In our review article, we provide researchers with the necessary understanding of how to change the dynamics of the cells to enable an analog operating mode.”

“You see time and again that groups simulate their memristor circuits with models that don’t take into account high dynamics of the cells at all. These circuits will never work.” said Stephan Menzel, who leads modeling activities at the Peter Grünberg Institute and has developed powerful compact models that are now in the public domain (www.emrl.de/jart.html). “In our review article, we provide the basics that are extremely helpful for a correct use of our compact models.”

Roadmap neuromorphic computing

The “Roadmap of Neuromorphic Computing and Engineering”, which was published in May 2022, shows how neuromorphic computing can help to reduce the enormous energy consumption of IT globally. In it, researchers from the Peter Grünberg Institute (PGI-7), together with leading experts in the field, have compiled the various technological possibilities, computational approaches, learning algorithms and fields of application. 

According to the study, applications in the field of artificial intelligence, such as pattern recognition or speech recognition, are likely to benefit in a special way from the use of neuromorphic hardware. This is because they are based – much more so than classical numerical computing operations – on the shifting of large amounts of data. Memristive cells make it possible to process these gigantic data sets directly in memory without transporting them back and forth between processor and memory. This could reduce the energy efficiency of artificial neural networks by orders of magnitude.

Memristive cells can also be interconnected to form high-density matrices that enable neural networks to learn locally. This so-called edge computing thus shifts computations from the data center to the factory floor, the vehicle, or the home of people in need of care. Thus, monitoring and controlling processes or initiating rescue measures can be done without sending data via a cloud. “This achieves two things at the same time: you save energy, and at the same time, personal data and data relevant to security remain on site,” says Prof. Dittmann, who played a key role in creating the roadmap as editor.

Here’s a link to and a citation for the ‘roadmap’,

2022 roadmap on neuromorphic computing and engineering by Dennis V Christensen, Regina Dittmann, Bernabe Linares-Barranco, Abu Sebastian, Manuel Le Gallo, Andrea Redaelli, Stefan Slesazeck, Thomas Mikolajick, Sabina Spiga, Stephan Menzel, Ilia Valov, Gianluca Milano, Carlo Ricciardi, Shi-Jun Liang, Feng Miao, Mario Lanza, Tyler J Quill, Scott T Keene, Alberto Salleo, Julie Grollier, Danijela Marković, Alice Mizrahi, Peng Yao, J Joshua Yang, Giacomo Indiveri, John Paul Strachan, Suman Datta, Elisa Vianello, Alexandre Valentian, Johannes Feldmann, Xuan Li, Wolfram H P Pernice, Harish Bhaskaran, Steve Furber, Emre Neftci, Franz Scherr, Wolfgang Maass, Srikanth Ramaswamy, Jonathan Tapson, Priyadarshini Panda, Youngeun Kim, Gouhei Tanaka, Simon Thorpe, Chiara Bartolozzi, Thomas A Cleland, Christoph Posch, ShihChii Liu, Gabriella Panuccio, Mufti Mahmud, Arnab Neelim Mazumder, Morteza Hosseini, Tinoosh Mohsenin, Elisa Donati, Silvia Tolu, Roberto Galeazzi, Martin Ejsing Christensen, Sune Holm, Daniele Ielmini and N Pryds. Neuromorphic Computing and Engineering , Volume 2, Number 2 DOI: 10.1088/2634-4386/ac4a83 20 May 2022 • © 2022 The Author(s)

This paper is open access.

Here’s the most recent paper,

Nanoionic memristive phenomena in metal oxides: the valence change mechanism by Regina Dittmann, Stephan Menzel & Rainer Waser. Advances in Physics
Volume 70, 2021 – Issue 2 Pages 155-349 DOI: https://doi.org/10.1080/00018732.2022.2084006 Published online: 06 Aug 2022

This paper is behind a paywall.