Error correction boost for AI memristor array
Researchers in South Korea have developed a neuromorphic AI processor using memristors that can both learn and process data.
The team at KAIST developed the chip with a 32 x 32 crossbar array with a highly stable memristor design using interfacial-type titanium oxide memristors with a gradual oxygen distribution that exhibit high reliability, high linearity and self-rectification. This allows for on chip error correction in the peripheral circuitry, linked to a digital controller to run AI algorithms in the analogue domain by self-calibration without compensation operations or pretraining.
Many teams in the US and Europe are developing memristor devices and architectures for AI< using the memristor array to store the neural network weights with more power efficiency than silicon memories, However the devices can be relatively unstable, reducing the accuracy over time.
- First complete memristor network for medical AI
- TDK, CEA team for AI spin memristors to cut power consumption
- TetraMEM memristors for AI in memory compute
The key advance at KAIST is that the chip can learn and correct the errors in the memristor array that occur due to non-ideal characteristics that were difficult to solve in existing neuromorphic devices. For example, when processing a video stream, the chip learns to automatically separate a moving object from the background, and it becomes better at this task over time.
The chip is ready for various devices including smart security cameras, allowing them to recognize suspicious activity immediately without having to rely on remote cloud servers, and medical devices that can help analyze health data in real time.
The low noise of average peak signal-to-noise ratio of 30.49 dB and self-learning ability has shown to achieve accuracy comparable to ideal computer simulations in real-time image processing.
Figure 1. Scanning electron microscope (SEM) image of a computing chip equipped with a highly reliable selector-less 32×32 memristor crossbar array (left). Hardware system developed for real-time artificial intelligence implementation (right). >
“This system is like a smart workspace where everything is within arm’s reach instead of having to go back and forth between desks and file cabinets,” said KAIST researchers Hakcheon Jeong and Seungjae Han, who led the development of this technology. “This is similar to the way our brain processes information, where everything is processed efficiently at once at one spot.”
The research was supported by the Next-Generation Intelligent Semiconductor Technology Development Project, Excellent New Researcher Project, PIM AI Semiconductor Core Technology Development Project, and the Electronics and Telecommunications Research Institute Research and Development Support Project of the National Research Foundation of Korea.
