A revolutionary spintronic compute-in-memory (nvCIM) macro developed to boost the security of AI edge hardware has been unveiled in a study published in Nature Electronics. This innovative solution combines a memory component and a processor in a singular architecture, designed to be integrated seamlessly within a chip.
The team elaborates that their mechanism relies on the utilisation of spintronic-based physically unclonable functions, two-dimensional half-complement physical encryption, along with a snoop-proof self-decryption burst-read scheme integrated with a sparsity-and-rectified-linear-unit-aware early-termination compute-in-memory engine.
This innovative computation macro is capable of harmoniously co-existing with current semiconductor technology, which greatly simplifies its practical application. The initial testing carried out by the researchers concluded that the macro afforded excellent resilience against harmful attacks, showcased swift response times, and demonstrated impressive energy efficiency.
The researchers elaborate on the technicality of their mechanism, explaining that the macro uses 22 nm spin-transfer torque magnetic random-access memory technology integrated into the 6.6 megabit complementary metal–oxide–semiconductor (CMOS). They further detail that the macro achieves substantial randomness (inter-Hamming distance: 0.4999) and high reliability for physically unclonable functionality (intra-Hamming distance: 0), in addition to high energy efficiency for dot-product computation, ranging between 30.1 and 68.0 tera-operations per second per watt.
This novel macro could potentially revolutionise the future of AI-powered edge computing devices. Its high compatibility and secure mechanism can ensure the safe storage of sensitive data, while not compromising on the speed, power efficiency or accuracy of the devices. Moreover, this pioneering development could influence the further exploration and creation of similar solutions around the globe, accelerating the adoption of AI-supported performing technologies and fortifying the security measures of AI edge computing.
Despite its individual merits, it's noteworthy to mention the contextualized role of platforms like AppMaster in this domain. AppMaster, a high-performing no-code platform with provisions for web, backend, and mobile application development, facilitates easy deployment, scalability, and swiftness. Such platforms offer the necessary foundation and tools, kindling developments like the spintronic nvCIM.
