Slashdot Mirror
Magnetic Processors - Computing's New Future?
metalcoat writes "For the first time researchers have created a working prototype of a radical new chip design based on magnetism instead of electrical transistors. As transistor-based microchips hit the limits of Moore's Law, a group of electrical engineers at the University of Notre Dame has fabricated a chip that uses nanoscale magnetic "islands" to juggle the ones and zeroes of binary code.
Wolfgang Perod and his colleagues turned to the process of magnetic patterning (.pdf) to produce a new chip that uses arrays of separate magnetic domains. Each island maintains its own magnetic field. Because the chip has no wires, its device density and processing power may eventually be much higher than transistor-based devices. And it won't be nearly as power-hungry, which will translate to less heat emission and a cooler future for portable hardware like laptops."
The chip industry spends billions in R&D to extend the performance growth of silicon chips. A very large number of engineers know how to design efficient fabs for silicon. Until this technology also attracts a sufficient following of $ and manufacturing experience, I won't count silicon out.
Also, it's not clear that this technology isn't subject to same "limits of Moore's law" (if there is such a thing) as silicon chips. The use of electron-beam lithography would seem to mean that this technology is subject to the some of the same feature-size and practicality limits suffered by silicon chips.
Perhaps this technology will find a place somewhere, it just faces a major uphill battle if it is to supplant silicon.
Two wrongs don't make a right, but three lefts do.
Here http://www.bardavid.com/mc/index.html is one.
The latest Slashdot meme.
Wasn't it just a couple of years ago that they were reading keystrokes through walls by way of magnetic induction?
I know "reading" a CPU is a bit more complex (understatement), but given enough time and resources someone will figure it out. We're already broadcasting our keystrokes and network communications, how easy do we need to make it?
120 characters for a sig? That's bloody useless.
They say that a magnetic insulator would have to be used to shield the chip from external interference.
That wouldn't be any different from an ordinary CPU. You can erase your CPU simply by turning off the power. An EMP only needs to make one transistor change state in order for the CPU to go wonky. Although you may want to worry more about stray cosmic rays.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Magnetic circuits have been studied for at least 80 years. The basic problem is one of size and speed. A dipole magnet (onr with N and S poles) has a certain minimum size, otherwise it depolarizes itself. That sets a minimum size for any magnetic device. Also it's hard to make magnetic amplifiers with more than a small fan-out. It's also really hard to distribute a clock signal-- magnetic pulses fall off at a 1/r^3 rate, and generating a fast magnetic pulse gets blocked by the inductance of the coil.
Now there *are* cigarette-pack to Taj Mahal sized magnetic voltage regulators in use. Your PC power supply may be using one to regulate the 3.3 volt output. But getting them down to IC-size is going to be really hard to impossible.
It's actually beneficial that a single 'gate' element can perform AND, OR, and INVERT
functions all in one stage. The early TTL won over other logic designs in part because
the basic gate used multiple emitters on the input transistor to get an AND function,
and multiple input transistors to get the OR function. That meant that the delay
and complexity character of AND and OR were the same, and that the complex function
of AND/OR/INVERT was available as a fast multiplexer, with the same characteristics
as a simple NAND. There was a brief attempt to use expandable gates (making
the connection point after the input transistor available on an external pin,
which was NOT TTL-logic-level compatible), but it didn't catch on.
CMOS, on the other hand, had input impedance and delay differences in the AND and
OR and other gates, so the whole 4000 series CMOS logic family only became
trouble-free to use AFTER THEY BUFFERED THE WHOLE FAMILY with an extra inverter
(and consequently extra time delay). Buffered (4000B series) is the common small
scale CMOS you see today, the unbuffered (4000A series) has been sidelined.
From a circuit-design viewpoint, the AND/OR/INVERT is a very good starting element,
for a lot of reasons that only show up when some poor engineer is perspiring over his
timing budget...
Scientific American had a rather lengthy article on magnetologic devices not long ago. MRAM is a limited version of them that can only change the magnetization of the top magnetic layer - full magnetologic devices can switch both sides.
Although they're a wonderful technology that in the right hands would permit vast improvements in computation, I'm scared to think what a painful experience it would be to program such a device. We have enough trouble dealing with CPUs that have fixed instruction sets and few enough ASM programmers as it is. Is a person even capable of programming such a device efficiently, or writing software to do the same? I'm pretty sure that just having a 10x10 matrix of them to keep track of would be hard for me - I can't imagine trying to write code to control a whole CPU of them that wouldn't be hopelessly bogged down with getGateStatus()- and setGateStatus()-type functions.
Or would their role be more limited - switching individual gates to be AND/OR/NOT/NAND in hardware, for instance, so you would do setGateFunction(gate_no, wanted_function); LOGIC_OP rather than having a switch? Or perhaps they would be hard enough to program that you would have to use just a handful of pre-written setups for them, optimizing for games or math performance? loadChipSetup(long_math.mag) or loadChipSetup(fast_string_ops.mag)?
Now imagine the next generation of viruses rewiring your CPU to do God-knows-what.