Slashdot Mirror
'Pruned' Microchips Twice As Fast and Efficient
Zothecula writes "If you had to use a commuting bicycle in a race, you would probably set about removing the kickstand, fenders, racks and lights to make the thing as fast and efficient as possible. When engineers at Houston's Rice University are developing small, fast, energy-efficient chips for use in devices like hearing aids, it turns out they do pretty much the same thing. The removal of portions of circuits that aren't essential to the task at hand is known as 'probabilistic pruning,' and it results in chips that are twice as fast, use half the power, and are half the size of conventional chips."
It's news that removing unnecessary parts of a circuit make it more efficient? Really?
Pruning. Go figure
I'll be removing the training wheels off my Harley this afternoon... thanks to this article I can be badass and efficient
Well DUH....
I was trying to make it more efficient by getting rid of some of the unused cores, so I got a pair of scissors and pruned off a couple of those cores. I put the pruned, aero dynamic chip back in my machine and now it won't start up! On the plus side, the power savings are noticeable :)
Why not just go with ASIC design if you don't need those fancy floating point units?
Someone's going to chime in and say that the naysayers are oversimplifying or denigrating this because they didn't think of it, but I think the quote below says enough.
Uh, no, Professor, I don't believe it is.
... for a specific application, like a hearing aid. Not so good for microprocessors intended for general purpose use (broad markets).
If you have sufficient market volume, you can afford to produce some sort of 'application specific integrate circuit'. Hmm, an ASIC. Now there's a novel idea (putting on jacket to make a dash to the patent office).
Have gnu, will travel.
From Wikipedia entry on Madman Muntz:
Ha ha!
"I believe this is the first time someone has taken an integrated circuit and said, 'Let's get rid of the part that we don't need,'"
I believe this to be a basic part of design.
Great. It's more efficient. Now they can charge even more, and your insurance won't cover it because it's experimental, but even if they do, you'll only get one hearing aid ever decade because your insurance deems that acceptable.
All good engineers do it, really...
I'm hearing impaired from birth (23 yrs). Just got my newest pair last week (previous pair is 5 years old, but working perfectly).
In several ways, this new pair is an upgrade.... but in one key way, I fucking hate these things. Both the previous and current hearing aids are digital (my previous pairs were analog). With the older pair there is a two second delay between turning the hearing aid on and hearing stuff. The new pair has a minimum of 6 seconds...
So if I need to scratch the inside of my ear quickly, I can do that in a second. Then I wait another 6 before I can hear again. Similarly, if I'm working without hearing aids in (relaxing, comfortable etc) and someone says something to me, I now have to wait almost 10 seconds before I can have them repeat what was said, then reply.
See the problem?
Hearing aid engineers are doing a lot of this work wrong (or for the wrong market, aka old people). Battery life is fine (a bit more than 2 weeks). I don't use the shitty auto-background-blocking programs on the hearing aid, either.
Yeah, I'm not that crazy about that idea.
From what I gather the components being removed are most likely resistors and capacitors. And sure, some can be probably removed, if you don't mind ending up with a noisy power supply and too much current going to various parts.
So you're left with a device that kind of works, but that may mysteriously stop working in a few months.
If you look carefully at hearing aid technology of the last 40 years, you'll see that digitization has brought precisely *nothing* to the benefit of the hearing impaired. It's still limited by the terrible quality mini-microphones, the very strange temporal smearing introducted by gain controls, and the "if everybody claps their hands together and really wishes hard, we can really tune this hearing aid perfectly!" game that is played by most audiology technicians, who have no scientific or experiemental basis for the settings they select. They have undocumented and local voodoo procedures for setting them.
Digitally sampling it and overprocessing the signal into "power bands" and handling them independently actually throws out critical timing information, that the nerves of the inner ear do process *if* they ever get the data. Simple direct amplification, and clipping excessively loud sounds, was shown to work well by Licklider in World War II. His work is basically annoyed by modern hearing aid manufacturers, all looking for the next patent and the next feature to get the families and insurance compoanies to invest in what is basically no better than an electronic ear trumpet, and which due to its undersized microphones and circuitry is swamped by thermal and environmental noise.
Given that, they can throw out the whole chip and replace it with, oh, about 50 transistors and make the batteries last 10 times as long.
He was probably responsible for those TVs I had when I was young which would lose sync when conditions weren't perfect, as in: either the Sun or the Moon were up.
Great minds think alike; fools seldom differ.
From my knowledge of integrated circuits (I am an EE), these small "unnecessary" circuits are generally used to improve stability and reliability. Yes, you can cut them out of an IC, but you run the risk of making a product that is significantly less reliable due to variations in the manufacturing process.
For example, it is possible to make an amplifier out of MOSFETs using only one p-channel and one n-channel, but typically this is not done because manufacturing processes can cause significant variation in the gain of an amplifier using such a configuration. For this reason, additional circuitry is added in order to make the amplifier more invariant to process variations. Granted, this is more true for analog circuitry than for digital, but the effect is similar for digital circuitry as well.
Between the time to boot and the "shitty" background blocking programs, I'm guessing that your new pair have some craptastic software.
That said, my hearing aids are almost ten years old. Even though digital, they don't have any noticeable lag between being turned on and being useful. But, on the other hand, my condition is genetic and my mother, 3 uncles and an aunt wear heariing aids. Many of them have very recent models and none have a six second lag. That's a pretty long span of time.
What an advance! If they continue on this path, they may even discover ASICs.
If I had to use a commuter bike that I could modify on a race, I'd be thinking about changing the gear ratio before dropping a marginal amount of weight.
Heroscape, it's like legos combined with anachronistic wargames.
also they are twice as good at doing half the nothing. They run empty infinite loops at half the power too.
You can't handle the truth.
The power drain for most hearing aids is measured in milli-amperes per hour. Moreover, many do not even have "off" switches.
"Pruning" has been around a while. Intel's been doing it since the 486 sx. That was just a laser zapped cpu that didn't use all of its components to get the job done at a less capable pace. You had to pay a premium to get the full DX. Now we're cutting things out to get the better performance. So now we will have to pay more for performance and for a lesser supply of materials. See, Capitalism wins once again. That's probably the real innovation here.
Patient I'm losing my hearing
Doctor (After checking him for anything serious) Don't worry it's part of the aging process.
Patient Can you do anything for me?
Doctor We can get you a hearing aid.
Patient How much will that cost?
Doctor Well, we have two models. One costs $1000 and the other costs $5.
Patient What's the difference between the two.
Doctor One is a sophisticated minitiarized amplifier assembled, customized to fit your ear, and tuned by highly trained technicians. The other is a button with a wire that goes to a wooden box in your pocket.
Patient What does the button do?
Doctor Nothing. But you'd be surprised at how much louder people talk when they see you wearing it.
For all intensive purposes, "whom" is no longer a word. That begs the question, "who cares"?
I say we call this a Reduced Instruction Set Computer! This could be BIG!
Is that so wrong?
Yes, you are right. This is the spirit of RISC.
But RISC was wrong. RISC resulted from a study of what instruction were actually used by typical applications that were compiled with standard compilers. This is like studying what railroad tracks are used and concluding that rail travel would be optimum if certain tracks were eliminated and others improved. This conclusion is wrong, because it assumes that existing rails include all optimal paths. In actuality, there might be paths that do not currently have rails.
Thus, the conclusion that a CPU should have fewer instructions is specious. A more accurate conclusion would have been that a CPU should have an optimum set of instructions for its intended task.
For example, consider the fact that linked lists are heavily used by most C programs. Yet, the C language does not have a linked list primitive: one has to use a library. Therefore, if a CPU had linked list operations built in (as the VAX did), a C compiler could not even use those operations because the language does not support it. Instead of concluding that the CPU's instruction should be fewer, one might conclude that the CPU should have linked list operations built in, and that linked list operations should be added to C. The result might be much faster programs.
My point is that RISC exposed the issue of the matching of CPU instruction to software, but the conclusion that CPUs should be simpler was wrong. The right conclusion would have been that CPUs should have optimal instructions, which might mean removing some and adding others - not merely removing some.
But you're off your rocker if you think that manufacturers are going to give device state information to the average joe that owns a hearing aid.
There is obviously more to the technology, but the "unnecessary" information has been pruned in order to make the article tighter and more accessible to the masses. Unfortunately, they removed all the bits that separate the approach from the engineering norm so it no longer functions as News for Nerds.
The key part they are removing is error detection and correction. They are creating chips which have an ~8% chance of producing an incorrect result. Supposedly hearing aids will accept a 10% error rate, so it is a good trade off.
These aren't "redundant" parts, they're parts which prevent errors from happening. It's just that in some applications they don't care about errors.
It's like looking at the various floating point bugs and going, "meh, close enough". Sucks for a spreadsheet, but if all you care about is integers 0-10, you probably aren't going to notice.
Here's the actual press release:
http://www.media.rice.edu/media/NewsBot.asp?MODE=VIEW&ID=15497&SnID=154992879
"Inexact Hardware" seems to be the new term. Since they mention hearing aids, it seems to be that it's bringing the fuzziness of analog back into the digital world.
this is the result of engineers who don't know how to design.
if you are _designing_ something, why in the world are you going to put in something you don't need ?
I spend my days trying to come up with clever ideas to NOT use things. Simple as possible and no simpler.
Absolute statements are never true
What you didn't mention is that "Muntz admitted his business lost $1,457,000 from April to August 1953,[28] and although he tried to reorganize, Muntz TV filed bankruptcy and went out of business in 1959" (from the same Wikipedia article)
You see, engineers don't sprinkle components at random. Every component in an electronic circuit is there for a reason. If something can be removed, what you have is a defective specification, maybe your circuit is designed to perform a function that's not often used, maybe it's designed to function in a situation that never happens. In that case you can ask the engineer to redesign for looser specifications.
Removing components at random is just stupid.
Perhaps the news is that enough time has passed since RISC that the notion is new again. Except with one sad and sorry exception. The RISC guys are way better engineers because they calculate various effects of their instruction sets and other optimizations BEFORE committing to silicon. The guys in TFA are relative chimps because they're just pruning, testing, and bandaging something they don't seem to fully understand.
An analogy comes to mind: some kids buy a car and pull off pieces until it doesn't work right. Once it breaks, they add duct tape until the car runs again. The stripped down car gets better mileage and can go faster. Somewhere along the way, some journalist looks at the thing and gets excited by the brand new car the kids built. The journalist trumpets to the sky about the wonderful new car and automotive geniuses he's discovered. The kids start a car company, dumb people invest. Meanwhile, automotive engineers furrow their brows, shake their heads, and go back to work.
I am a lawyer, but not yours. Anything I tell you might be a total lie intended to benefit my clients at your expense.
Sounds a bit like how Woz built the first Apple computers, finding ways to do more with less.
comment first, facts later. http://chem.tufts.edu/AnswersInScience/RelativityofWrong.htm
The FPU was an expensive low-yield section of the circuit on the 486 processor. Often, the 486sx parts were ones with defects in the FPU section, so they just disabled the FPU and sold them cheap.
Removing parts from a chip that are not required for the operation of said chip results in a chip that is smaller, faster, etc.
In other news, scientists announce that water is wet.
This reminds my of the Pareto Principle (http://en.wikipedia.org/wiki/Pareto_principle) . Named after an Italian economist who observed something similar in 1906.
I get the feeling that this is one of those ideas that we will continue to discover on a periodic basis.
Most of the logic in a chip (ASIC or CPU based design) is not visible as user features. I would guess probably 40% of the a silicon can be removed without the end customer knowing about it. That doesn't mean those transistors are not needed.Here are some examples of "non user visible" circuits
- SCAN (for testing the flops) can take up to 20% of the area
- Redundant memories (2 or more bits per memory location) to increase yield at a specific speed
- Extra logic to account for process variations
- Debug blocks
- Manufacturing visibility blocks
These days most of the Moor's law improvements is used to reduce power and increase manufacturing efficiency.
I don't think Timothy is a nerd. I don't even think he knows what one is.
I'm sorry if this offends you tim, but from what i've seen, you are really fucking lame. You post up the stupidest fucking articles possible.
What, does your 8 year old cousin do this for you?
I understand the concept of "practice makes perfect" but you should realize that there is some things we aren't good at, and this job you do, here at slashdot, isn't for you. You suck at it. You haven't gotten better, you've gotten worse. Why don't you find something else that's more your speed. I don't know what that is, and I don't care. As long as it is far away from slashdot.
Be seeing you...
SPEED HOLES!!!!!
Digital chips are roughly comprised of memory (flip flops) with logic in between. On each clock cycle the logic takes data from one piece of (input) memory, transforms it in some way and stores it in some other (output) memory.
One of the primary limitations on the speed of the chip is the longest path. The length of a path is roughly a function of the physical length of the path that the data takes from input to output memory and the number/type of logic gates in between. The speed of the chip is roughly limited to 1/[Time taken for data to cross the longest path] Hz
If you remove a significant amount of logic there is a chance that you might be removing the longest and most complex paths for data to cross and also that all the components can be spaced closer together, meaning that there is less distance for the data to cover. For this reason, removing parts of the chip might be able to speed it up.
http://freefall.purrsia.com/ff1900/fv01807.htm
Laptop password related links: dell bios password dell primary password dell latitude bios password dell inspiron bios password dell xps bios password dell vostro bios password dell hard drive password acer bios password hp bios password laptop bios password
Laptop password related links: dell bios password
A little more full featured than a RISC (Reduced Instruction Set Computing) processor and less bloated than current X86? Sounds like what should have been happening already.
grandma just got fitted for a new pair ~$6400, they will last 4 years. Spendy!