Can a 3D chip change AI forever? It can move data faster, uses less energy, and packs more power in less space. Find out more!
Modern AI chips need to move large amounts of data between memory, which stores information, and the parts that process it. On flat chips, memory is spread out, so data has to travel long paths. The processors work faster than data can arrive and there is not enough memory nearby, so the chip slows down. This is called the memory wall. For years, chipmakers tried to fix this by making transistors smaller and putting more on the chip, but this is reaching its limits, called the miniaturization wall.
Researchers from Stanford, Carnegie Mellon, the University of Pennsylvania, and MIT worked with SkyWater Technology to create a multilayer computer chip. Its design could help AI hardware and domestic semiconductor production.
The new 3D chip integrates memory and computation vertically. The design allows faster information transfer, like elevator banks in a building move people simultaneously. By combining memory and logic and building upward, the chip can fit more computing power into a smaller space.
Early tests show the prototype outperforms 2D chips by about four times. Simulations of taller designs with more layers show up to twelve-fold improvements on AI workloads. The design also allows possible 100- to 1,000-fold gains in energy-delay products, a metric that balances speed and energy use. By reducing data movement and adding many vertical pathways, the chip delivers higher throughput with lower energy per operation, a result hard to achieve with flat designs.
This work sets a blueprint for domestic semiconductor production. By showing monolithic 3D chips can be made in U.S. foundries, it opens the way for hardware to be designed and built domestically. Moving to vertical 3D integration will need engineers trained in these methods. Educational and research programs are preparing students for this next wave of U.S. semiconductor work.
The research represents not only performance gains but also added capability. 3D chips could speed innovation, enable faster responses to challenges, and shape the future of AI hardware.
