Still, there were some clear challenges with MOS.2. In normal silicon, the threshold voltage of the transistor can be adjusted by doping with silicon. This is an implant impurity that changes the behavior of a semiconductor. However, there is no way to implant impurities into a single molecule. All RV32-WUJI semiconductors are N-type and cannot be adjusted for their performance. Therefore, the researchers here used two different metals (aluminum and gold) for the wiring, adjusting the threshold voltage of each transistor through wiring selection, and adjusting the material with embedded wiring.
Make a chip
At the chip level, researchers experimented with building many individual devices and used machine learning to identify the optimal combination of wiring and materials that ensure that each transistor is present within the required performance envelope.
At the transistor level, the device uses what is called Depletion Mode Inverter. To build functional circuits, researchers built and tested a complete suite of 25 logic gates and tested them. Eight people were functional, and researchers used them to build chips. They used the longest path through the chip to determine the delays that must be explained and set an upper limit to clock speeds in the kilohatsu range. The overall yield when we finally made the chip was above 99.9%, with chip-level yield of 99.8%.
That said, some of the circuits have proved to be quite challenging. For example, the yield of 8-bit registers was only 71%, dropping to just 7% for 64-bit registers (requires 1,152 transistors).
The resulting processor contains 5,900 individual transistors, allowing the entire 32-bit version of the RISC-V instruction set to be implemented. This necessarily means that it includes advanced circuits such as the RISC-V instruction decoder. At the same time, some aspects are intentionally kept simple. You can add two 32-bit numbers, but do so by manipulating one bit at a time. This means that it takes 32 clock cycles to perform the operation. This also required an onboard buffer to store intermediate results.
Still, it works and the authors claim it is perhaps one of the most refined bits of hardware “beyond silicon.” That said, they don’t expect this technology to replace silicon. Instead, they believe it potentially meets several niche needs, such as ultra-low power handling of simple sensors. However, as technology continues to advance, the range of potential applications could expand beyond that.
Nature, 2025. doi: 10.1038/s41586-025-08759-9 (About DOI).
