Chinese researchers have developed a flexible artificial intelligence (AI) chip capable of enduring more than 40,000 bending cycles without affecting performance. The chip maintained its performance and stability throughout a 6-month trial period, according to a study published in Nature.
FLEXI, the chip is thin, roughly 25 micrometers, or about one-third the thickness of a standard A4 sheet of paper, and lightweight. In lab testing, it withstood over 40,000 cycles of 180° bending without any impact on performance and remained stable through continuous trial operation over six months, according to the report.
According to the published paper, FLEXI achieves clock frequency operation up to 12.5 MHz and power consumption as low as 2.52 mW. The design incorporates co-optimization across process technology, circuit elements, and algorithmic deployment.
Flexible AI chips that can bend are an emerging frontier in semiconductor research worldwide. Unlike traditional rigid silicon processors, these tiny devices are designed to operate under mechanical stress and support edge intelligence in wearables, robotics, and medical electronics. Many global technology companies and startups are developing AI processors optimized for edge, low-power, or diverse form factors.
In flexible AI chip electronics research, bend endurance is commonly reported in the range of 20,000-30,000. Crossing 40,000 bending cycles places a device in the top tier globally, especially when the bending angle is severe (≈180°). Many flexible circuits fail mechanically or electrically well before this threshold.
The chip can bend to tight radii approaching 1 millimeter while maintaining stable electrical performance, a key requirement for real-world wearable and embedded applications.
The chip’s smallest version measures about 31.12 mm² and integrates 10,628 transistors, yet can operate in an ultra-low power mode, consuming just 55.94 microwatts.
According to the Nature journal paper, the researchers behind the FLEXI flexible AI chip have identified several specific application areas where this technology could be useful.
1. Wearable Health-Monitoring Devices
FLEXI isn’t just mechanically resilient; it’s also capable of credible AI inference, In trials reported in the Nature paper, FLEXI, was tested on physiological signals such as heart rhythm, and the chip achieved up to ~99.2% accuracy in detecting cardiac arrhythmias in controlled experiments. This demonstrates its potential to form the core processing unit of low-cost, long-duration wearable medical monitors that can analyze biosignals locally without heavy reliance on smartphones or cloud servers.
2. Human Daily Activity Monitoring
The Nature paper reported that FLEXI’s on-chip neural networks were able to classify routine activities — such as walking or cycling, with high accuracy (>97 %), showing that the flexible AI architecture can do more than basic sensing. This positions the chip for human activity recognition (HAR) in smart wearables and fitness technologies.
3. Smart Textiles & Conformal Wearables
Because FLEXI’s flexible substrate allows it to physically conform to the body or curved surfaces, the researchers suggest it could serve as the compute engine in next-generation smart textiles, garments and fabric-based devices embedded with sensing and AI computation.
4. Flexible Brain Computer Interfaces
CGTN mentions that the chip’s flexibility and performance under repeated bending make it suitable for brain–computer interface (BCI) applications.
The research was led by interdisciplinary teams from Tsinghua University and Peking University. Experts say flexible AI chips like FLEXI could unlock a wave of applications in smart textiles, medical wearables, impulsive robotics, and even conformal human-machine interfaces, domains where traditional rigid electronics have struggled to gain traction.
The true significance of a flexible AI chip is not its ability to bend 40,000 times, but where intelligence can finally live. FLEXI shifts AI from rigid enclosures into environments that are dynamic, organic, and continuously in motion.
Also Read: Micron Launches $24B Singapore Chip Plant to Meet Surging AI Demand
