Published in the prestigious journal Nature Electronics, the breakthrough achieves analog computation accuracy comparable to digital systems<\/strong>, improving analog precision by five orders of magnitude<\/strong> \u2014 or nearly 100,000 times<\/strong>.<\/p>\n This achievement marks a monumental step in post-Moore era computing<\/strong>, where energy efficiency and processing power matter more than transistor counts.<\/p>\n 5 Key Takeaways<\/p>\n The 100-Year Challenge: Why Analog Computing Lagged Behind<\/p>\n Analog computing<\/a> is not new. In fact, it predates digital computers. Early analog machines could model differential equations and simulate physics problems long before silicon processors existed.<\/p>\n Yet, analog computing faced a fatal flaw \u2014 inaccuracy<\/strong>. Noise, variability, and drift made analog systems unreliable for large-scale, precise computations. As digital computing <\/a>advanced, analog methods faded into history.<\/p>\n But with the explosion of AI, 6G, and edge computing<\/strong>, the demand for faster and more energy-efficient processing<\/strong> has reignited interest in analog computing \u2014 and Peking University\u2019s team may have just solved its biggest weakness.<\/p>\n techovedas.com\/huawei-recognized-as-digital-for-life-champion-in-singapore<\/a><\/p>\n Inside the Breakthrough: The RRAM-Based Analog Matrix Chip<\/p>\n The Peking University researchers created a resistive random-access memory (RRAM)<\/a><\/strong> chip that performs matrix-based analog computation<\/strong> with unprecedented accuracy.<\/p>\n RRAM technology allows data to be stored and processed in the same physical location \u2014 a concept known as in-memory computing<\/strong>. <\/p>\n Unlike digital chips that move data back and forth between memory and processor cores (a process that consumes huge energy), RRAM executes computations directly within memory cells.<\/p>\n By fine-tuning the resistance states of RRAM and developing precision calibration algorithms, the team achieved digital-level accuracy<\/strong> \u2014 something previously thought impossible for analog systems.<\/p>\n \u201cThis is the first time analog computation accuracy has rivaled that of digital systems,\u201d said Sun Zhong<\/strong>, lead researcher. \u201cWe improved traditional analog precision by nearly 100,000 times.\u201d<\/p>\n<\/blockquote>\n techovedas.com\/what-emerging-memories-types-and-advantages<\/a><\/p>\n Performance Beyond GPUs<\/p>\n The performance results are staggering.<\/p>\n When tested on large-scale MIMO (multiple-input multiple-output) signal detection<\/strong> \u2014 a key task in advanced communication systems \u2014 the analog RRAM chip demonstrated:<\/p>\n In simple terms, the chip can process complex computations faster and with far less power, making it ideal for next-generation technologies that demand real-time processing.<\/p>\n A Leap Toward Post-Moore Era Computing<\/p>\n The semiconductor industry has long depended on Moore\u2019s Law<\/strong>, <\/a>which predicts transistor density doubling every two years. But as transistor sizes approach atomic limits, the digital performance gains that once fueled innovation are slowing.<\/p>\n This has pushed global researchers to explore new computing paradigms<\/strong> \u2014 quantum computing, neuromorphic systems, and now, high-precision analog computing.<\/p>\n Peking University\u2019s analog chip could become a cornerstone of post-Moore computing<\/strong>, where progress is defined not by smaller transistors, but by smarter architectures and lower energy footprints<\/strong>.<\/p>\n \/techovedas.com\/what-is-moores-law-more-than-moore-and-beyond-moore<\/a><\/p>\n Real-World Applications<\/p>\n The implications of this research extend far beyond the lab.<\/p>\n 1. 6G Communications<\/p>\n Next-generation 6G networks<\/strong> will require base stations to process massive antenna signals in real time. 2. Artificial Intelligence Acceleration<\/p>\n AI training relies on heavy matrix multiplications. The analog chip can accelerate second-order optimization algorithms<\/strong>, drastically improving training efficiency<\/strong> for large models. 3. Edge and On-Device AI<\/p>\n Perhaps the most transformative application lies in edge computing<\/strong>.<\/a> techovedas.com\/made-in-india-vaaman-reconfigurable-power-redefines-edge-computing<\/a><\/p>\n Why It Matters: A New Frontier for China\u2019s Semiconductor Innovation<\/p>\n This breakthrough also underscores China\u2019s growing strength in semiconductor R&D<\/strong>.<\/a> Developing a chip that can outperform digital GPUs in specific workloads \u2014 using homegrown materials, design, and algorithms<\/strong> \u2014 marks a strategic leap<\/strong> toward technological independence.<\/p>\n It reflects a broader trend in China\u2019s research landscape: moving from following global standards<\/strong> to setting them<\/strong>.<\/p>\n techovedas.com\/top-semiconductor-manufacturing-countries-in-2025-whos-leading-the-global-chip-race\/<\/a><\/p>\n Expert Opinions: A Global Turning Point<\/p>\n Global semiconductor experts have hailed the development as a landmark moment<\/strong> in computing history.<\/p>\n An editorial in Nature Electronics described it as \u201ca fundamental step toward merging analog and digital computation at scale.\u201d<\/p>\n Industry analysts also note that this could reshape AI hardware, offering an energy-efficient alternative to <\/strong>Nvidia\u2019s GPUs<\/strong> and TPUs<\/strong><\/a> in specific applications like inference acceleration and wireless signal processing.<\/p>\n \u00a0NVIDIA Crowns OpenAI King of AI with World\u2019s First DGX H200 \u2013 techovedas<\/a><\/p>\n Challenges Ahead<\/p>\n Despite the optimism, scaling this innovation to commercial use remains challenging.<\/p>\n Still, the Peking University team is confident that this is the start of a new era of analog-digital fusion<\/strong> computing.<\/p>\n\n
![\"\"](\"https:\/\/www.europesays.com\/us\/wp-content\/uploads\/2025\/10\/image-91-1024x576.png\")
<\/p>\n\n
\n
This analog RRAM chip can handle such large-scale signal detection tasks with ultra-low power<\/strong>, boosting network efficiency<\/strong> and data throughput<\/strong>.
That means faster internet speeds, lower latency, and greener connectivity.<\/p>\n
In essence, it could reduce the training time and energy cost of systems like GPT models or image recognition AI<\/strong> by massive margins.<\/p>\n
With its low-power design, this analog chip enables AI inference and training directly on devices<\/strong> \u2014 without relying heavily on the cloud.
That means smarter, more autonomous devices capable of learning and adapting locally \u2014 from smartphones and drones to medical wearables and autonomous cars.<\/p>\n
As the U.S. and its allies impose restrictions on advanced chip exports, Chinese researchers are pivoting toward alternative computing architectures<\/strong> instead of catching up in digital silicon alone.<\/p>\n![\"\"](\"https:\/\/www.europesays.com\/us\/wp-content\/uploads\/2025\/10\/image-92-1024x576.png\")
<\/p>\n\n
RRAM devices can suffer from variability across large arrays, affecting precision at scale.<\/li>\n
Merging analog chips into existing digital infrastructures will require hybrid architectures<\/strong> and new design frameworks.<\/li>\n
Today\u2019s AI and computing frameworks \u2014 like TensorFlow and PyTorch \u2014 are built for digital logic. Bridging them to analog hardware needs new software-hardware co-design<\/strong> efforts.<\/li>\n<\/ol>\n