In a bid to overcome shortcomings in scientific computing, Chinese scientists have unveiled a new approach to sorting data that promises both higher speed and lower energy consumption. The system combines memristors—electronic components that mimic the memory function of the human brain—with an advanced sorting algorithm to process large amounts of information more efficiently.
Researchers say this method could help overcome performance bottlenecks in not just computing but also artificial intelligence (AI), and hardware design, where rapidly organizing and analyzing vast datasets is essential. Beyond AI, potential applications for this technology include smart traffic systems that analyze images in real time and financial services that require quick risk assessments.
Prototype shows memristor sorting boosts route finding and neural inference
To demonstrate the potential of their technology, scientists from Peking University and the Chinese Institute for Brain Research created a hardware sorting prototype based on memristors. The system successfully handled tasks like route finding and neural network inference, delivering faster performance and lower energy consumption compared to traditional sorting methods, the South China Morning Post reported.
Overall, the system achieved a 7.7-fold increase in throughput and improved energy efficiency by more than 160 times compared to conventional sorting methods. It also boosted area efficiency by over 32 times, marking a significant step towards integrating storage and computing for broader, general-purpose applications.
In a paper published in Nature Electronics last month, the team explained that sorting remains a major performance limitation across applications ranging from artificial intelligence and databases to web search and scientific computing.
Traditional computing systems rely on the Von Neumann architecture, which separates data storage and processing functions, typically using a central processing unit (CPU) to handle calculations.
New insights into memristors show potential to revolutionize computing
According to the researchers of the latest study, the conventional system has led to the Von Neumann bottleneck, which limits the speed of data transfer between memory and processing units. They explained that while sort-in-memory approaches using memristors could help overcome these limitations, current systems still depend on comparison operations, keeping sorting performance constrained.
Unlike ordinary resistors, which simply reduce the flow of electricity in a circuit, memristors have the unique ability to remember how much electrical charge has passed through them. This memory function allows memristors to adjust their resistance based on previous activity, enabling them to act as both storage and processing components.
By combining these functions, memristors could eliminate the need to transfer data between separate memory and processing units, potentially leading to faster and more energy-efficient computing systems.
The scientific team aimed to simplify sorting by removing the need for comparison units. Traditional hardware sorting relies on CPUs, GPUs, or specialised chips that compare numbers step by step using sorting algorithms. Instead, the new method uses memristors to perform iterative search-based sorting, finding minimum or maximum values without directly comparing each pair, which in turn saves both time and energy.