D5 Advanced Modern Electronics

Topic

The cutting edge of electronics is advancing in several simultaneous directions. Chips specialized for artificial intelligence execute—in massive parallel—a very specific and repetitive type of arithmetic operation (multiply-and-accumulate) that forms the basis of nearly all deep learning algorithms; in doing so, they achieve performance and energy efficiency far superior to those of a general-purpose processor performing the same task. They represent, in essence, the same trade-off between flexibility and efficiency seen in the FPGA-versus-ASIC comparison, pushed to the extreme of a chip designed for a single type of operation.

Application-specific circuits, designed from the ground up to perform one highly specific function, can be orders of magnitude more efficient than any programmable alternative for that particular task—though this comes at the cost of a design and manufacturing investment justified only by very high production volumes. Quantum computing relies on qubits built from superconducting circuits as its basic element; these must be kept at extraordinarily low temperatures—fractions of a degree above absolute zero—to prevent thermal noise (a phenomenon described by models of electronic and quantum electronic noise) from triggering unwanted spontaneous transitions in their delicate quantum states.

In recent decades, the semiconductor industry has split into companies that design chips without owning their own factories and specialized manufacturing facilities—known as foundries—that actually produce the chips using the most advanced manufacturing processes available. The manufacturing yield of any chip—the fraction of chips produced on a wafer that are functional—decreases as the area of ​​each individual chip increases and as the defect density of the manufacturing process rises. That is why this module—which opens with chips designed for a single type of operation—concludes by showing that the frontier of electronics advances not by inventing new physics, but by pushing the very trade-offs established in this catalog to their limits: efficiency versus flexibility, thermal noise versus quantum stability, and chip area versus manufacturing yield—each one advancing a little further than before.

ID:4076