Driftspace
A chip designed to run artificial intelligence tasks while consuming around 98 p
A chip designed to run artificial intelligence tasks while consuming around 98 percent less energy than comparable conventional processors has been presented by Japanese automotive-parts supplier Aisin, working with Tohoku University. Researchers say the central innovation is embedding memory directly within the chip using magnetic RAM technology, eliminating the energy-intensive step of shuttling data to and from separate external memory modules — typically one of the largest power costs in AI processing. The findings were presented at an international semiconductor conference held in Taiwan this month. If the efficiency gains prove achievable at larger scales, the implications could extend well beyond automotive applications: AI data centres, medical edge devices and consumer electronics all face mounting pressure to reduce power use as AI workloads expand globally. The company has indicated that exploring a path toward commercial production is its next step. As concerns about AI's total energy appetite grow, could a hardware breakthrough like this change how you think about artificial intelligence's real-world costs?