Saturday 23 November 2013
Computer engineering, but in particular mobile computer engineering, is all about playing a zero-sum game with yourself. Power and efficiency are constantly undercutting one another, creating confounding incentives for designers looking to set records for both talk time and processing speed. At this point it seems obvious that both speed and battery life are limited by the old process of laying down increasingly dense little fields of silicon transistors; whether it’s a quantum computer or a graphene chip, getting more computing power for less electrical power will require a fundamental shift in how we build computers.
A new 
study from UC Berkeley hopes to provide the basis for just such an advance, laying out their attempt at a silicon replacement they say uses up to 10,000 timesless power than prior solutions. They have designed a system that uses magnetic switches in place of transistors, negating the need for a constant electric current. The idea of a magnetic transistor has been discussed since the early 1990s, but the idea’s downfall has always been the need to create a strong magnetic field to orient the magnets for easy switching; all or most of the power saved by the magnets is spent creating the field needed to actually use those magnets.
study from UC Berkeley hopes to provide the basis for just such an advance, laying out their attempt at a silicon replacement they say uses up to 10,000 timesless power than prior solutions. They have designed a system that uses magnetic switches in place of transistors, negating the need for a constant electric current. The idea of a magnetic transistor has been discussed since the early 1990s, but the idea’s downfall has always been the need to create a strong magnetic field to orient the magnets for easy switching; all or most of the power saved by the magnets is spent creating the field needed to actually use those magnets.
This new study, published last week in Nature, uses a wire made of tantalum, a somewhat rare element used to make capacitors in everything from Blu-Ray players to mobile phones. Tantalum is a good, light-weight conductor, but it has one particularly odd property that’s made it uniquely useful for magnetic applications: when a current flows through the tantalum wire, all clockwise-spinning electrons migrate to one side of the wire, all counter-clockwise-spinning to the other. The physical movement of these electrons creates a polarization in the system — the same sort of polarization prior researchers have had to create with an expensive magnetic field.
If this approach were successful and practical, we could begin to capitalize on some of the shared benefits of all magnetic computing strategies, the most glaring of which is that magnetic switches do not require constant current to maintain their state. Much like a liquid crystal in an e-ink display, a magnetic transistor will maintain its assigned state until actively flipped. This means that a theoretical magnetic processor could use far less energy than semi-conducting silicon ones by accruing energy savings whenever it is not actively doing work. And since tantalum is a fairly well-known material, its incorporation into the manufacturing process shouldn’t prove too difficult.
Raw tantalum.
One interesting thing about this ability to maintain an assigned state is that it essentially makes the chip itself programmable. Where silicon transistors must be laid down physically for each specific function, magnetic switches could be reoriented by software to meet a specific need. Decoding video is a very different process than rendering out that same video in a real-time graphics engine, and the two processes use distinct arrangements of physical transistors. A magnetic chip could theoretically change its arrangement on the fly, presumably in response to software commands, to better suit itself to the particular task at hand.
So, magnetic transistors offer a way out of the zero sum game in which increased power necessarily hacks off battery life, and decreased power consumption requires a slower overall speed. The real-world power of a chip made of magnetic-switches will of course be limited not by science but by manufacturing — making experimental transistors is nice, but ultimately meaningless if we can’t churn out thousands of such chips on relatively short notice.
Though distinct in many ways, these transistors still use the same basic on-off logic as regular transistors, so they’d need a comparable production standard to compete in terms of raw speed. My Nexus 5 boasts a 28-nanometer Snapdragon chip that packs four high-speed cores onto a chip the size of a graham cracker; though magnetism has its advantages, it’s likely that silicon will continue to reign for a very long time to come.
- By Graham Templeton on November 20, 2013 at 2:14 pm
