Researchers at the Technion and the Tower Jazz company have developed a technology that adapts commercial transistors to the age of artificial intelligence

The research published in the journal Nature Electronics was led by PhD student Louis Daniel and Prof. Shahar Kutinsky from the Viterbi Faculty of Electrical Engineering at the Technion, in collaboration with Prof. Yaakov Roizin and Dr. Yevgeny Pihai from the Tower Jazz Company and Prof. Mishna Ramez Daniel from the faculty for biomedical engineering at the Technion.

Researchers at the Technion and the Tower Jazz company have developed a technology that turns Tower Jazz's commercial flash memory components into memristors - devices that contain memory and computing power. The technology, which was developed inspired by the workings of the human brain, significantly accelerates the operation of artificial intelligence (AI) algorithms.

The research published in the journal Nature Electronics was led by PhD student Louis Daniel and Prof. Shahar Kutinsky from the Viterbi Faculty of Electrical Engineering at the Technion, in collaboration with Prof. Yaakov Roizin and Dr. Yevgeny Pihai from the Tower Jazz Company and Prof. Mishna Ramez Daniel from the faculty for biomedical engineering at the Technion.

From the very beginning of their journey, computers surpassed humans in solving invoice problems, but in recognizing images, classifying characteristics within the image, and making decisions, the computer lagged behind humans for decades. This gap has been caught up in recent years by artificial intelligence, which is able to perform complex operations based on training based on examples. In recent decades, enormous resources have been devoted to the development of artificial intelligence at the software level, an investment that has led to a leap forward in the effectiveness of artificial intelligence in many diverse fields, including medicine, smart transportation, robotics and agriculture.

The fuel that drives the world of artificial intelligence is data, and more specifically data in huge amounts (big data). This is why the big breakthrough in artificial intelligence "waited" for the dramatic improvement in computing power. But the rapid development in software performance has left the hardware behind, and the development of hardware suitable for the requirements of artificial intelligence software has been delayed for many years. Such hardware is required to work well in terms of speed, low power, accuracy, space and price. All of these are very difficult to achieve in the traditional hardware model - a model of digital (numerical) calculation.

The digital model limits hardware performance in two main contexts: 1. Such hardware has difficulty performing many operations at the same time, since they are designed to perform a relatively small number of operations; 2. They can demonstrate great accuracy only in return for consuming very large resources in terms of energy and time. Therefore, the researchers say, innovative hardware is required to fit the age of artificial intelligence.

"One of the major challenges that artificial intelligence poses to hardware engineers," explains Prof. Kotinsky, "is the realization of complex algorithms that require (a) storage of a lot of information in the computer's memory; (b) rapid retrieval from memory; (c) performing many calculations at the same time; and (d) high accuracy. The standard digital hardware (processors) is not suitable for this for the reasons stated above."

This is the background to the new technology presented in the article in Nature Electronics. According to Prof. Kotinsky, "Our technology turns the hardware, which is essentially digital, into a neuromorphic infrastructure - a quasi-analog infrastructure similar to the brain. As the brain performs millions of operations at the same time, our hardware also performs many operations at the same time and thus speeds up all operations related to it."

"The neuromorphic calculation," says PhD student Louis Daniel, "interests me on a personal level both as a computer engineering student and as someone who lost his father as a result of a rare neurological disease. The brain has always been a source of inspiration for computing systems, and my challenge is to understand the computational mechanism of the brain's operation through an engineering toolbox. In the current study, we showed an electrical chip based on commercial standard technology and demonstrating two critical capabilities: associative memory, which works similar to the brain based on features rather than searching in an index; and ability to learn.”

The associative memory, familiar to us from the act of human thinking, means that when we see eyes, for example, we are not looking for a match of the eye to some section in an index of items, but identify the eye associatively. It is a fast, effective and energy-efficient mechanism. And as in the brain, the learning ability of the system is improved by changing and updating the connections between the synapses and the nerve cells.

According to Prof. Ruizin of Tower Jazz, "the new technology is simple to implement and turns the Tower Jazz transistor, which was originally designed to store information only, into a memristor - a unit that contains not only memory but also the ability to calculate. Since the meristor sits on top of the existing TowerJazz transistors, it instantly interfaces with all the devices that these transistors work with. The new technology was tested under real conditions and showed that it can indeed be implemented in the construction of neural networks in hardware and thus significantly improves the performance of commercial artificial intelligence systems. Similar to the brain, the improved system excels in long-term information retention and very low energy consumption."

According to Associate Professor Ramez Daniel, formerly an electrical engineer at Tower Jazz and currently a faculty member in the Faculty of Biomedical Engineering at the Technion, "the computational power of the improved device stems from its ability to function in the field of subconduction, and in simple words - in a manner similar to natural biological mechanisms. As a result, we achieve high efficiency at low power, like the mechanisms that have developed in nature over billions of years of evolution."

Technion researchers Eric Herblin, Nicholas Weinstein, Vasu Gupta and Nimrod Wald from Prof. Kotinsky's research group participated in the study.

The research was conducted with the support of the National Planning and Budgeting Committee, a Kamin grant from the Ministry of Economy, the Andrew and Erna Finci Viterbi scholarship for graduate students and an ERC grant. Louis Daniel recently presented the above research at the Nature conference in China and even won the award for the outstanding poster article at the conference.

About the research partners:

Prof. Shahar Kotinsky completed a bachelor's and master's degree at the Hebrew University and a doctorate at the Technion and worked at Intel in circuit design. After a post-doctorate at Stanford University, he returned to the Technion as a faculty member in the Viterbi Faculty of Electrical Engineering. Over the years, he has won many awards, including the Wolf Foundation's Krill Award for excellence in scientific research, the Viterbi Scholarship, the Jacobs Scholarship and the ERC grant, as well as seven teaching excellence awards.

Louis Daniel completed his bachelor's degree at the Technion, and in 2016-2013 worked at IBM's research laboratories in Haifa. He is currently doing his doctorate (on a direct track) under the supervision of Prof. Kotinsky. He won the Herschel Rich Prize for Innovation and Entrepreneurship, the Andrew and Erna Finci Viterbi Prize for graduate students, and the VTA Scholarship for outstanding doctoral students from Arab society.

Prof. Yaakov Roizin is the director of research and development of futuristic technologies and a Fellow at Tower-Jazz. He is a visiting professor at the Technion and Tel Aviv University. Yaakov has extensive experience of 40 years in the field of component development in the semiconductor industry. For the past 23 years he has been working at Tower Jazz and specializes in the development of unique CMOS technologies and innovative components. He has published more than 200 studies and holds more than 50 patents (USA patents) in the field of semiconductor and component technology.

Dr. Yevgeni Pihai is a senior component engineer at Tower Jazz. He has 15 years of experience in the development of CMOS components that include integrated memories (embedded NVM), solar cells and ionizing radiation sensors. Yevgeni completed a bachelor's degree at the Technion, a master's degree at Tel Aviv University and a doctorate at the Technion. He has published over 40 articles and patents.

Associate Professor Ramez Daniel completed a bachelor's degree at the Viterbi Faculty of Electrical Engineering at the Technion and a master's degree in electronics and electrical engineering at Tel Aviv University, after which he went into industry. After eight years of work at Tower Jazz, he went on to do a PhD followed by a post-doc at MIT, where he built the first biological computer inside a bacterium. He is currently the head of the Synthetic Biology Laboratory at the Faculty of Biomedical Engineering at the Technion.