Slashdot Mirror
Electromechanical Switches Could Reduce Future Computers' Cooling Needs
Earthquake Retrofit writes "Science Daily is reporting that researchers at Case Western Reserve University have taken the first step to building a computer capable of operating in extreme heat. Te-Hao Lee, Swarup Bhunia and Mehran Mehregany have made electromechanical switches — building blocks of circuits — that can take twice the heat that would render electronic transistors useless. 'The group used electron beam lithography and sulfur hexafluoride gas to etch the switches, just a few hundred nanometers in size, out of silicon carbide. The result is a switch that has no discernable leakage and no loss of power in testing at 500 degrees Celsius. A pair of switches were used to make an inverter, which was able to switch on and off 500,000 times per second, performing computation each cycle. The switches, however, began to break down after 2 billion cycles and in a manner the researchers do not yet fully understand. ... Whether they can reach the point of competing with faster transistors for office and home and even supercomputing, remains to be seen. The researchers point out that with the ability to handle much higher heat, the need for costly and space-consuming cooling systems would be eliminated.'"
Miniaturized relays are interesting, but an inverter which operates at 0.0005 Ghz is less interesting. Somehow I don't think we'll be seeing this replace electronics anytime soon. (well, except in lithium battery microcontrollers :-) ). Although it would be interesting technology for a steampunk novel.
Hundreds of nanometers is rather larger then the current tech 32nm. These are going to have to get quite a bit smaller, faster, and more durable before they stand a chance. A two billion cycle limit is the worst part i think, my cpu did that in the last .66 seconds.
I tried to tell them that tubes and transistors were just a fad. Relays were good enough for the Z4 and they're good enough for us. These kids and their newfangled gadgets...
Warning: this article may contain humor, sarcasm, parody, and perhaps even irony. Read at your own risk.
Both attributes that the military would like.
... is easily explained. There is a flaw in the design. Someone apparently made the "wear out expiration" register signed instead of unsigned.
now we need to go OSS in diesel cars
There are two sources of heat in modern semiconductor CPU's.
One is leakage, the heat generated by current times resistance squared in transistors that are off.
The higher current that is related to the clock speed is the heat generated by transistors that are turned on using the same current times resistance squared.
To keep the on current at a bare minimum, transistors are paired with one on and one off so the current through the pair should be zero except for leakage. The current flows when they are clocked and the capacitance (stored voltage) of the wire between transistors and gate capacitance of the MOSFET it drives supplies current during switching.
How does this no heat switch avoid the current of switching the capacitance between the switches. From what I can tell is this part is able to handle higher temperatures. I do not see it as a no power (no heat generated) device.
Silicon Nitride has much higher resistance than most metals. Due to the resistance and temperature resistance, it is often used as hot surface ignition in gas appliances. Current through the switches will create heat. It is unavoidable.
At it's current speed of 0.000.5 GHZ clock speed, I can believe the current power consumption is very low. How does this stack up to an Atom CPU clocked at 0.000.5 GHZ?
The truth shall set you free!
This is a research idea that MAY be useful, the demise of CMOS silicon has been highly exaggerated.
From the summary:
"an inverter, which was able to switch on and off 500,000 times per second" -> 500kHz is not so great
"however, began to break down after 2 billion cycles" or about 1 second at current processor speeds. That increases to 4000 seconds at 500kHz, or a little more than an hour.
Also, we can put billions of error free transistors on a chip for a few dollars. THAT is the real hurdle that nothing else has been able to clear yet. We will likely be with silicon for a while after it stops shrinking for this reason.
Nobody's going to use this for desktop CPUs. The whole point is that the switches work at 500 degrees C, where silicon doesn't. This technology would be used for embedded control in extremely hostile environments, where 500 kHz would be just fine. The article names the inside of a jet engine and the surface of Venus as examples.
Visit the
We had the opposite problem.
The sprinkler activation system was disconnected at all times (excepting inspection). That made me feel safe.
Eventually -I- figured out what was wrong with the system and fixed it, and with the alarms cleared they felt safe enough to put the activators back on.
For large sets, this will be our guide even unto death, for the LORD will work for each type of data it is applied to...
I do test circuit hardware design and we use standard relays all over the board, for switching bits of circuitry into and out of contact with an integrated circuit we're testing. We use mechanical relays because of the same reasons they say: zero leakage current when they're open, and extremely low resistance when they're closed, which semiconductor switches just can't equal. The problem is the lifetime of the relays, so we have to socket them all (which, when you're building a board with 500 relays on it, is a significant time and money sink) and replace them pretty often on high-running parts (some of our parts have been in high-volume production for 20 years.) Plus they're big and take up the majority of the board. Having a device that's tiny and can last a billion cycles would be completely awesome.
Nostalgia's not what it used to be.
The venerable 6502 was also the heart of the VIC 20. A slightly modified/improved version was used in what was the most ubiquitous personal computer, the Commodore 64. Although running at only 1Mhz, most of its instructions executed in less than 3 clock cycles, making for some pretty efficient and fast ML code.
Willie...
First we have flash memory can that only be written to N number of times, and now they're building a cpu that can only do N computations?
Seems like something interesting for planetary exploration where standard CPUs on a probe would be rendered useless in a matter of hours. Much as the equipment sent to Venus.