The human brain is the most complex and capable computer man has ever known. This marvel of nature can simultaneously process millions of data points and sensory input in real time. While computers are now capable of surpassing the human brain’s capabilities in terms of computing power, there are still some areas in which they lack. A recent breakthrough in memristors could hold the key to unlocking neural network computing, opening the door for smarter, faster artificial intelligence systems in the future.
Current Chip Design
Today’s von Neumann computing architecture places the processing and memory in two separate locations. Originally, this layout made sense, as computers didn’t require the sheer amount of data retrieval found today. However, advanced AI systems process billions of bits of data in real-time. As such, every millisecond counts, including the time and energy it takes to retrieve and process memory from two separate locations.
Problems with Current Chip Design
The current chip design is too slow to keep up with the growing demands of the AI community. A better solution for computational heavy protocols like AI systems is to place the memory and processing on a single chip. To accomplish this task, engineers turned towards a technology created in the 1970s, memristors.
Memristors
The concept of Memristors originated in 1971 following a paper published by
Leon Chua. Memristors function through the use of varying currents. Changing the direction and amount of current enables these devices to store data. Their compact size and ability to store and process data, provide them with ideal performance for today’s advanced AI systems.
Problems with Memristor Tech Today
The main issue with today’s memristor tech used in computers is that they can’t compete with other systems yet. These experimental chip designs often provide lower computational power, questionable reliability, and lack the durability of their dual-chip counterparts. Thankfully, all of that is about to change.
Neuromorphic Computing
Memristors are a crucial component of the neuromorphic computing movement. This computer class seeks to mimic the neural networks found in your brain. These systems can compute, process, and store data in a single location. They can also learn, correct, and adjust to new scenarios, just like how human brains operate. This capability, coupled with the new inclusive chip design enables a new era in computational processing power.
Memristor Chip Study
The study “Self-supervised video processing with self-calibration on an analog computing platform based on a selector-less memristor array” was published in the scientific journal Nature Electronics this month. The study details a next-gen neuromorphic semiconductor-based processor that mimics the way the human brain thinks, enabling it to learn and adapt to improve performance.
New Chip Design
The researchers integrated a selector-less analog memristor array to create an all-in-one storage and processing unit. The device leverages peripheral circuitry and a digital controller to program the memristors in real time. Specifically, interfacial-type titanium oxide memristors were chosen to run the experiments.
Source – KAIST
Mimics the Human Brain
One of the most impressive parts of this research is that the engineers were able to integrate an advanced AI system directly into the chip. This integration allows the chip to learn and correct errors. The device provides real-time analysis and onsite processing, making it the ideal option for multiple use-case scenarios like music editing, security, and much more.
Figure Out New Scenarios
Impressively, the engineer chose to run AI algorithms in the analog domain. Consequently, the system was set up to utilize self-calibration without compensation operations or pretraining. Like a person, this AI can evaluate the real-time data and cross-reference it to its history to determine the next action.
Improve its Performance
This setup allows engineers to create computers and devices that learn, correct errors, and process AI tasks onsite. The data is then stored and processed to see its relation to past data and other unique features. Whenever the system determines that performance improvements are possible from learning, it automatically integrates them.
Memristor Chip Test
In order for the engineers to test the world’s first memristor-based integrated system that can adapt to real-time data changes, they set up a video editing task. The AI system was fed continuous real-time video of a moving person. It was tasked with separating the foreground and background of the video.
Memristor Chip Test Results
The data shows some promising results. For one, the AI got faster and more efficient with each cycle. At first, the AI took extra time to determine the situation, however, on the final runs through, the AI had already become accustomed to the process and how to determine what is back or foreground.
Impressively, the AI chip video editing test showed an average peak signal-to-noise ratio of 30.49 dB, comparable to dedicated services. Additionally, the team registered a structural similarity index measure of 0.81.
Memristor Chip Benefits
There are several reasons why memristor chips are a smart idea. For one, they offer onsite real-time data analysis. This means that devices can operate without internet connections or the need to ping another device to make an educated assessment of its environment.
Security
Security and privacy are increased thanks to this layout. Since the data doesn’t need to be sent offsite, it greatly reduces the risks of hacks or human error.
Better Performance
Another major benefit of this study is that it demonstrates how to improve performance from today’s PCs. The memristor-powered AI system offers parallel computation capabilities to its users, improving capabilities and enabling systems to handle more load.
Size
Memristor chips are much smaller than traditional layouts. As such, they are ideal for microelectronics, wearables, robotics, and other scenarios where space is limited or weight is a major concern, like drones or other aerospace tech.
Reduced Energy Consumption
In terms of energy consumption, memristor chips utilize hundreds of times less power than traditional chip design. The energy used to program the chip is miniscule. This low power consumption is ideal for sustainability and ensuring that the AI revolution doesn’t take a huge toll on the environment like the industrial revolution continues to do.
Memristors Applications
There are many applications for memristors in the market. For one, they are the ideal options for compact and energy-efficient artificial intelligence (AI) edge-computing systems. This technology could one day help to make all of your smart devices much smarter, lighter, and more capable. Here are just a few applications for the memristor chip system.
Video Processing
AI video processing is one of the fastest-growing sectors in the market. The researcher’s chip is ideal for this task for many reasons. For one, it uses its self-learning capabilities to reduce its workload and improve its output. You could see this style of chip used to power advanced AI editing systems like the one tested by the team.
Already, AI editing systems are making viewing more affordable. Three systems often can create duplicate frames between the original frame, enabling massive resolution upgrades. In The future, a system like this could make it simple to edit high-end video from your phone.
Healthcare
The use of AI in healthcare is on the rise. Memristor-powered smart devices will one day provide in-depth monitoring of patients, empowering professionals by lowering their workload and improving their awareness. The AI could even be set up to send the data to another AI system designed specifically for healthcare.
Security
Using a self-learning AI system to program security cameras makes sense. Imagine your smart camera monitoring your yard. Suddenly, a person appears and makes a few movements before walking off-screen. If you weren’t watching the screen, you missed it. A memristors-powered AI security system could scan the area for suspicious behavior and notify you in real time.
Autonomous Vehicles
Autonomous vehicles are sure to integrate this technology in the future to enable them to take on new responsibilities and a higher level of reliability. The system will allow vehicles to analyze their environment and make educated guesses as to the best scenario. Every Time, the decision is evaluated and used to measure the next option, ensuring better performance over time.
Memristors Researchers
This study was led by KAIST engineers Professor Shinhyun Choi and Professor Young-Gyu Yoon’s Joint Research Team from the School of Electrical Engineering. Hakcheon Jeong and Seung Jae Han also did their part to gather and process the data.
The memristor chips study has support from the Next-Generation Intelligent Semiconductor Technology Development Project, Excellent New Researcher Project and PIM AI Semiconductor Core Technology Development Project of the National Research Foundation of Korea, and the Electronics and Telecommunications Research Institute Research and Development Support Project
A Company that Could benefit from the Memristor Study
There are many industries this technology could improve. The world has reached its limit in terms of shrinking resistors. This latest development introduces a more reliable and scalable option to the market. Here’s one company that is positioned to leverage this data for success.
Micron Technologies Inc. (MU -13.69%) entered the market in 1978. It was founded by Ward Parkinson, Joe Parkinson, Dennis Wilson, and Doug Pitman as a semiconductor manufacturing plant. The company quickly saw success and in 1980, the firm opened its first fabrication plant.
Micron Technology, Inc. (MU -13.69%)
Since then, Micron Technologies has expanded its offerings into other tech-related sectors. Impressively, It even secured a spot in the Fortune 500 in 1994. Today, it’s recognized as a pioneering force following decades of chip innovations.
Micron Technologies is a leading researcher and developer of memristor chipsets. The company seeks to integrate this tech to drive its thriving data center business model. Currently, the company employs 43,000 people. Consequently, MU is seen as a strong “buy” for traders seeking exposure to the AI sector.
Meristors will Power the AI revolution.
The innovative team behind memristor CPU design understands the importance of direct on-chip memory and processing in terms of powering AI systems Now, the engineers will further their research and testing to get memristor-powered chips into consumer electronics as fast as possible.
Learn About Other Computing Breakthroughs.
