Are big, cheap dies possible?

[Shalmanese]

Heres the thing, I have several friends who are computer pack-rats. They never throw anything away. Every winter, they get all their old computers cranking to keep their rooms at a nice toasty temperature. Since electricity turns to heat no matter what you do to it, they all figure that having it computer SETI or do a bit of raytracing is a good a use as any.

Now, I was wondering if it was possible to mass produce custom built computer/heaters. The main charecteristics, of course is that they need to be cheap and they need to heat well. Now, Silicon seems to fail once you get above 100C or so and there seems to be no way around that so the key to a large heat output is obviously larger die sizes. I know with modern CPU's, the trend has been towards smaller dies at lower costs but is there anything inherent about large dies that make them so expensive?

The proposed CPU would neccesarily need to have low transistor densities, wider gates and lower clock speeds otherwise the W/cm^2 would just be too much for any passive cooling system to cope. Does this make it cheap enough to produce large enough die for my proposed solution, or, alternatively, lots of smaller dies.

I'm not in any way expecting to be able to make $20 target heaters but something like a $1000 installation art/conversation piece heater would have a viable chance of making it IMHO. Is it possible?

 

[thelordemperor]

The silicon wafers that AMD and Intel buy make up the bulk of the production costs for new CPUs, so making cheap heaters out of CPU-grade silicon is probably not possible.

A simple electric coil heater is probably close to as efficient.

 

[iwantanewcomputer]

if you want heat use gas heating. i't way cheaper than electric. If you want computing power get a real computer. It costs intel/amd just as much to make the top model cpu as models that are over a year old. it even costs less(1-2%) to make 90 nm cpus than 130 nm cause the die size is smaller, and they can make more at a time. the major cost determinant is the technology to make a specific processor, but once they have that they might as well make p4 prescotts instead of p1's. the best way to make heaters that can do a bit of computing to is to get a dual-4way prescott based (nocona) xeon system. these produce as much heat as any processor and performance is nearly tops. if you want them to be chep at the expense of computing ability get old athlon MP's

 

[cy7878]

well, if this is for art and not really function. i suppose you can design a board to power up a bunch of old Pentium 60's, with the CPU and a passive heat sink (no fan) exposed outwards. The old P-60 burns up about 60 watts each. If you power-up about 10 of these, you got heat! Now you can probably find them about a buck a piece.

 

[CTho9305]

Originally posted by: cy7878
well, if this is for art and not really function. i suppose you can design a board to power up a bunch of old Pentium 60's, with the CPU and a passive heat sink (no fan) exposed outwards. The old P-60 burns up about 60 watts each. If you power-up about 10 of these, you got heat! Now you can probably find them about a buck a piece.

Given that a 75MHz Pentium's max power is 8 watts, I highly doubt a properly functioning Pentium 60 will use 60 watts.

 

[wfbberzerker]

if you want some heat, get a bunch of monitors.

 

[clarkey01]

Or make a fire ;-)

 

[Sunner]

Works wonders in my apartment.
8 computers, 5 21" monitors, and a couple of switches heats it up.

Unfortunately I don't like heat

 

[imported_jediknight]

In CPU manufacturing, the cost of making a chip is approximately proportional to the die size ^ 4 (I'm not sure what was factored into these calculations.. it was in my integrated circuits text :->) so I'd say no, large dies for the purposes of heating are not practical..

 

[Calin]

Originally posted by: CTho9305

Originally posted by: cy7878
well, if this is for art and not really function. i suppose you can design a board to power up a bunch of old Pentium 60's, with the CPU and a passive heat sink (no fan) exposed outwards. The old P-60 burns up about 60 watts each. If you power-up about 10 of these, you got heat! Now you can probably find them about a buck a piece.

Given that a 75MHz Pentium's max power is 8 watts, I highly doubt a properly functioning Pentium 60 will use 60 watts.

http://www.intel.com/support/processors/pentium/sb/cs-011039.htm

A Pentium 60 has a maximum power of 14.6W, while a P66 has 16W (with typical values of 11.9W and 13W)
The original Pentium processors had a higher operating voltage, this makes the difference between them and the faster P75

Calin

 

[mrSHEiK124]

and i was about to make my old P200MMX a keychain, now it can become a furnace! too bad you don't really need them when you a. live and florida and b. have central heating

 

[Shalmanese]

Originally posted by: jediknight
In CPU manufacturing, the cost of making a chip is approximately proportional to the die size ^ 4 (I'm not sure what was factored into these calculations.. it was in my integrated circuits text :->) so I'd say no, large dies for the purposes of heating are not practical..

That's a lot, any ideas why? I suspect it's something to do with yield and defects but making a processor that could re-route around defects could be an idea. I suspect with the current batch of high density, high speed, low die size processors, this wouldn't be very useful but it would if your going for large dies for the sake of large dyes.

 

[Eskimo]

Originally posted by: thelordemperor
The silicon wafers that AMD and Intel buy make up the bulk of the production costs for new CPUs, so making cheap heaters out of CPU-grade silicon is probably not possible.

Actually the bulk of the cost of creating a finished wafer is the overhead associated with all the process steps necessary to go from starting silicon to final passivation. There are equipment costs both in purchasing and maintenance as well as significant raw materials costs such as process gases and chemicals. The wafers themselves are lumped into the raw materials category but aren't in themselves an overwhelming piece of the pie. More critical for companies like Intel than the cost of the wafers is the availability of them. In the US at least labor costs are the second biggest factor, a major reason why more and more production is moving overseas.

A simple electric coil heater is probably close to as efficient.

Very true.

 

[Shalmanese]

In that case, shouldn't cost be largely independant of die size? after all, a 10 step process on 200mm^2 is not much different from a 10 step process on 400mm^2.

 

[Eskimo]

No, the cost is directly related because while your total cost to produce the finished wafer is a function of your process complexity the cost to produce a single die from that wafer is a function of the die size and yield rate. If it costs me $4000 to make a finished wafer and i can get 400 saleable die out of it my die cost is $10. If I can only get 40 working die then my cost is $100. Does that make sense to you?

The total number of die canidates available on the wafer is a function of your die size. The larger the die the less total canidates you can print on the wafer. Yield is a funciton of systematic loss due to process integration issues and random loss due to defectivity during processing. Your random yield loss (Yr) will increase exponentially with die size.

 

[Shalmanese]

Okay, in that case, wouldn't cost increase largely linearly with die size (discounting losses due to fitting square dies on a round wafer). OT: Has anyone experiemented with doing several different sized dies on a single wafer so as to maximise wafer usage? ie: big dies in the middle and then smaller ones around the edges.

 

[Shalmanese]

Okay, in that case, wouldn't cost increase largely linearly with die size (discounting losses due to fitting square dies on a round wafer). OT: Has anyone experiemented with doing several different sized dies on a single wafer so as to maximise wafer usage? ie: big dies in the middle and then smaller ones around the edges.

 

[jagec]

Originally posted by: cy7878
well, if this is for art and not really function. i suppose you can design a board to power up a bunch of old Pentium 60's, with the CPU and a passive heat sink (no fan) exposed outwards. The old P-60 burns up about 60 watts each. If you power-up about 10 of these, you got heat! Now you can probably find them about a buck a piece.

if its heat you want, go Cyrix

 

[Eskimo]

Originally posted by: Shalmanese
Okay, in that case, wouldn't cost increase largely linearly with die size (discounting losses due to fitting square dies on a round wafer).

Yes, cost to create a die increases linearly with die size. However as I mentioned you must also consider yield. The cost to create a die increases exponentially (inverse natural log) in terms of random yield loss due to random defectivity.

OT: Has anyone experiemented with doing several different sized dies on a single wafer so as to maximise wafer usage? ie: big dies in the middle and then smaller ones around the edges.

While i'm sure it has been tried there are two major sticking points I can think of to trying that.
A) Most die of different sizes are not the same product, different products usually require different processes which can make for a very complex flow. Take implants for example. In some cases your large die may need an implant that your small die does not, you would somehow have to mask off your small die and leave portions of your large die open. This leads to the second point.

2) Each sized die would require a different set of masks. It's not practical to switch reticles in the middle of an exposure of a wafer. Especially considering you are usually processing multiple wafers through a linked track/stepper, 25 for 200mm, 25 or 13 for 300mm. You would have to switch out reticles for each wafer OR you'd have to expose all of your large die on all the wafers, develop, then skip the post develop bake, hope your resist was still sensitive and send it through again this time exposing only your small die.

Summary: Very complex, adds more cost than the benefit of silicon savings.

 

[Shalmanese]

Ah, I see. I was under the impression that a single die was etched in one go. ie: the photomask contains n different copies of the same processor. If they do them 1 by 1, then I can see why the added expense is not worth the hassle.