Artificial intelligence has been rapidly advancing, with applications such as generative AI pushing the boundaries of existing digital hardware. The scalability of CPUs, GPUs, and ASICs is being tested to its limits, leading to a surge in research into analog hardware specialized for AI computation. Analog hardware offers the ability to adjust the resistance of semiconductors based on external voltage or current, utilizing a cross-point array structure with vertically crossed memory devices for parallel processing.
Despite the advantages analog hardware presents for specific computational tasks and continuous data processing, there are challenges in meeting the diverse requirements for computational learning and inference. To tackle these limitations, a research team led by Professor Seyoung Kim focused on Electrochemical Random Access Memory (ECRAM). ECRAM devices manage electrical conductivity through ion movement and concentration, featuring a three-terminal structure with separate paths for reading and writing data, enabling operation at relatively low power.
In their study, the research team successfully fabricated ECRAM devices in a 64×64 array using three-terminal-based semiconductors. Experiments demonstrated that the hardware incorporating these devices exhibited excellent electrical and switching characteristics, high yield, and uniformity. By applying the Tiki-Taka algorithm, an advanced analog-based learning algorithm, to this hardware, the team was able to maximize the accuracy of AI neural network training computations. The research also highlighted the impact of the “weight retention” property of hardware training on learning and confirmed that their technique does not overload artificial neural networks.
This research is groundbreaking as it represents the largest array of ECRAM devices for storing and processing analog signals to date. While previous literature only reported arrays up to 10×10, the research team successfully implemented these devices on a larger scale with varied characteristics for each device. The potential for commercializing this technology is evident, showcasing the promise of analog hardware using ECRAM devices in maximizing AI computational performance.
The research team’s findings demonstrate the significant potential of analog hardware with ECRAM devices for enhancing AI computational performance. By addressing the limitations of analog hardware memory devices and implementing advanced learning algorithms, the team has laid the foundation for commercializing this technology and revolutionizing the field of artificial intelligence.
Leave a Reply