#Nudging the user
21 messages in this thread
Brooks,
When I was working for AMD, there was a group working on the possibility of
integrating batteries along with the flip-flops in RAM chips. After the
Japanese pretty much took over the RAM market, that work stopped…but it was
an interesting concept: self-powered RAM. It's possible to extend the concept
to almost any other type of circuit. If all IC's incorporated the battery cells
to power them, then, with the exception of the need to perform an occasional
recharge, battery life would (presumably) be a lot longer.
I agree that the development of a revolutionary battery is of utmost
importance…maybe a very tiny nuclear reactor! <g>
-rich-
>> I agree that the development of a revolutionary battery is of utmost
importance…maybe a very tiny nuclear reactor! <g>
Actually, this will be an essential component of any Pentium-based subnotebook.
The only technical challenge is board placement of the cooling tower for
optimal heat dissipation.
Ha!
While I certainly hope this is a major stumbling block for the evil empire, I
don't put a lot of faith into Apple's marketing nostrums. Surely Intel has some
scheme for notebooks besides the aging (and duplicated) 486 line. Seems like
there has to be a future Pentium that won't have the power and heat problems of
the desktop versions.
Perhaps Motorola, in making the perfect chip, has taken too long to get it to
market. I hope not (I really don't want to learn Windows and have to develop
for systems where AI is required just to determine a file type….).
Pentium isn't living up to performance or yield expectations, while the 601 is
exceeding expectations on both fronts.
I don't doubt that Intel will manufacture a lower-power version of Pentium, but
then what?
What remains to be seen, and will be played out over the next couple of years,
is whether customers are so attached to their 80×86 code base that they are
willing to endure a mostly dead-end technology path at great expense. If they
are, that's too bad for them and disastrous for Apple.
Shifting to a PowerPC code base will be uncomfortable, but the hardware will be
cheaper and faster. And isn't that how DOS/Windows really came to dominate the
market anyway? It wasn't because DOS/Windows is a better front end, it's
because that happened to be the OS of choice on the fastest, cheapest hardware.
Have to agree with your comments (and begin _really_ wondering what started
this thread off before we all perverted it – nudging the user? huh?). No
matter – the only thing I would add (or subtract) is that DOS/Windows won out
not because it was cheaper and faster but simply because it was cheaper (and
maybe, initially, backed by IBM which used to mean something to businessmen).
Another factor is the multiple suppliers, which insured competition, which
meant not having to wait for hard drives to appear, etc.
Price of a CPU is determined primarily by development cost – a fixed expense.
It follows that the more chips you ship the cheaper the chips can become.
Pentium prices are probably artificially high because Intel knows that demand
will far outweigh initial supplies. If current Windows users take to Pentium
machines (is there any doubt?) the price will fall.
Given that, I hope you are right. I think customers are attatched to product
features and interface and that these attributes are rapidly becoming standard
across platforms, except in the operating systems arena. Its the OS that will
decide the winner.
Craig,
"Actually, this will be an essential component of any Pentium-based
subnotebook. The only technical challenge is board placement of the cooling
tower for optimal heat dissipation."
If you've looked at the PowerPC 601 chip's specs, you'll see that the Pentium
is a much lower power chip, by far, than the 601 (i.e., 16 W for the Pentium
versus 85 W for the 601)! The 603 chip is supposed to have a much lower power
consumption.
Surprised? I was.
-rich-
Rich,
==>
If you've looked at the PowerPC 601 chip's specs, you'll see that the Pentium
is a much lower power chip, by far, than the 601 (i.e., 16 W for the Pentium
versus 85 W for the 601)!
<==
Umm, that's 8.5 watts. Those damn decimal points!!! 🙂
The pentium may draw 16 W, which might not sound like a lot, but any chip with
it's own integrated fan is _way_ too hot IMO.
Besides, I heard from an inside source that Penium chips are being rejected at
a 30% rate during manufacturing. They're expensive and they're gonna stay that
way.
Cheers,
Rob
Robert,
"Umm, that's 8.5 watts. Those damn decimal points!!!"
You're absolutely right. I looked at the ad once again, using a magnifying
glass, and there was, indeed, a decimal point there. <Whew!>
That's a whole lot better. I couldn't figure out how Motorola managed to build
a package that could dissipate 85 watts. Thanks for the correction.
"I heard from an inside source that Penium chips are being rejected at a 30%
rate during manufacturing."
It's not unusual to have low yields, particularly with a new chip that uses new
technology. The recipe hasn't yet been tweaked so that they get consistently
good yields. When the 80286 came out, iNTEL was getting 2 wafers per die (yes,
that's right. Only one die in every two wafers was good). Fortunately, they
learned how to make the chip and got very respectible yields after a short
time. When I was working for AMD, we always had low yields on new devices.
Manufacturing LSI devices is really very much of a black art. It's not just a
simple matter…and even the slightest change in the gaseous diffusion stage or
metal deposition stage…or any other stage of a chip's manufacture, for that
matter, can cause an entire run to turn to s***.
The cost to manufacture any chip of equal size is the same. The high prices on
brand new chips is the manufacturer's attempt to cash in on the early sales
income, which can be considerable in the beginning. The 80286 chips were
selling for over $400 each, at first…but after six months or a year, and
particularly after AMD was giving them some competition, the prices dropped
dramatically. Back in the late 70's, it cost about $200/wafer to manufacture
_ANY_ device, regardless of its complexity (afterall, it doesn't cost any more
to photolithograph a complex device than a simple device). The difference
between complex (i.e., larger) devices and simple (i.e., smaller) devices is
that you can get more of the latter on a wafer. The cost for chemicals and
labor is almost identical on a per-wafer basis.
No doubt iNTEL (and Motorola) are going to cash in on the early demand for
their devices by charging premium prices (they charge whatever the market will
bear…and they do _lots_ of market surveys before setting the (relatively
arbitrary) price). The Pentium and the Moto 601 chips will decline about 15%
(or more) in price each year, depending upon sales volume (did they make enough
money up front?), yield (are they getting more good die/wafer now?), and
competition (is Moto or AMD beating down the price of their competitive
devices?).
Thanks again on the correction on the 601's power consumption. I was really
perplexed by what I read as 85 watts, when—as you pointed out—there's
really a decimal in there, making it 8.5 watts. That's a relief.
Thanks,
-rich-
Rich,
Is there any kind of generalization one could make on cost given the area of a
chip. For example the Pentium has around twice the area of a PowerPC 601. Is
it possible to roughly estimate expected manufacturing costs using todays tech
keeping everything but the chip size constant?
Later,
Rob
Robert,
"Is there any kind of generalization one could make on cost given the area of a
chip."
Yes and no. All manufacturing costs are figured on a per-wafer basis. Depending
on the size of the wafer, you might get more or less potential die. Depending
upon the average yield/wafer, the cost per good die can vary all over the
place. Because of the Pentium's size, it will be much more subject to possible
imperfections, and therefore (presumably) provide lower yields. Silicon wafers
are still not 100% pure crystals and any of the processing steps can introduce
flaws into the manufacturing step.
If you knew the "inside information" on the average cost per wafer for the
Pentium or PowerPC 601, you could figure out a cost per die, based upon the
wafer and die sizes. However, it's unlikely that you would be able to get the
necessary information.
Back in around 1980, it was estimated that manufacturing costs for _any_ kind
of chip, regardless of its complexity or size was about $200/wafer. For very
small chips, where you could get 100's on a wafer (e.g., TTL circuits), the
cost per chip was very low. For large chips, where only a few would fit on a
wafer (then, a 4" wafer size was fairly standard), both the potential for flaws
because of the chip's size and the fact of life that no wafer produces all good
chips, the cost per chip was fairly high (especially if you got only about 4-10
good chips per wafer).
I'm not well enough up to speed on the current semiconductor technology to give
you any kind of estimates. Early per-chip manufacturing costs are high, owing
to low yields, mainly. That's why the semiconductor manufacturers get premium
prices at the beginning. However, as time goes on, the process is "tweaked",
and yields rise, prices can be lowered…although a sole-source supplier of a
high-demand device is not too anxious to do so.
It's a very complex business and I don't think there's any way to estimate
manufacturing costs without being privy to all of the variables associated with
the manufacture of a given chip. You really have to be "inside" to know that
information, and it is usually very tightly held.
-rich-
Rich:
It's not just wafer size. One has to also consider the number of process steps.
The number of layers in the device also adds considerably. So perhaps a cost
per process step class is a way to get a better grip. For example the chemicals
for etching are indeed 'standard' costed but the number of repetitions is a
function of the number of layers. Similarly with deposition etc.
Then of course one has to factor in the number of wafers per boat. Plus what
about the cost of testing/grading. If it's statistical it's less than if it's
100 %.
Finally packging, plastic versus ceramic etc.
All in all a very complex calculation.
best
Peter
Peter,
I was obviously oversimplifying the situation, merely to offer a relatively
concise response. Of course, all the elements you mention contribute to the
cost of manufacturing a device. There is also any labor-intensive work, such as
wire bonding (that isn't automated), as well as the cost for testing (which for
complex devices such as CPU chips is a HUGE expense, especially in terms of the
manpower required to write the test programs). The labor costs probably greatly
outweigh the costs of the materials, I'd guess. Then, there is the depreciation
of the capital equipment needed for manufacture, which is nontrivial in
anyone's book.
The result is that if one delves deeply into trying to specify a cost for
manufacture of a given device, it becomes a terribly complex issue. My main
point was that, in general, manufacturing cost is relatively constant (in the
long view) and that manufacturers tend to get most of their profits from a new
device up-front, when it is in great demand and they can charge a premium for
it. The mitigating factors (yield, etc.) have already been covered.
Let's not get too involved in trying to enumerate all of the factors which
influence manufacturing cost. I'm sure that the cost accountants of the
semiconductor industry are making a good living doing just that. 🙂
-rich-
Rich,
I absolutely agrre with you…I was just surprised to discover you were from
the industry. (Plus you are into MIDI! small world to coin a phrase)
I used to work on CIM software for making semi-conductors and had the good
fortune to visit Japan
a number of times. Also Fishkill!
best
Peter
Peter,
"I was just surprised to discover you were from the industry."
I spent several years designing new MOS LSI devices, including the world's
first binary floating-point processor (the AMD Am9511). I enjoyed the
experience and learned a lot about the technology and also the "black art" of
semiconductor manufacturing. I've also had quite a bit of experience with AMD's
RISC processor technology (Am290x0 series devices). However, all of that is in
the past. I'm into other things now, and I'm sure that my information about
semiconductor manufacturing is way out of date by now.
I wish I had more time to play with my MIDI equipment. I seem to spend most of
my time writing programs and books. It's very enjoyable, but there are other
things I'd like to do but just can't seem to find time. <sigh>
Take care,
-rich-
Yes, I'm afraid you're mistaken.
I've seen a Pentium cook a hot dog at 16W.
No microprocessor has ever burned 85W, except that Z80 we plugged into line
voltage in sophomore EE lab. And it didn't burn 85W for very long.
Craig,
You're right. I was mistaken. My old eyes didn't see the decimal point in
Motorola's ad for the 601. I come from a semiconductor background and couldn't
imagine how Moto could dissipate 85 watts in an LSI package…but I believed it
until I was recently told to look for the decimal point, which, indeed, is
there.
The 601 will burn only 8.5 watts. <whew!>
-rich-
>>If you've looked at the PowerPC 601 chip's specs, you'll see that the Pentium
is a much lower power chip, by far, than the 601 (i.e., 16 W for the Pentium
versus 85 W for the 601)! The 603 chip is supposed to have a much lower power
consumption.<<<
NOT!!
Try 16 Watts for the Pentium and ONLY 8.5 watts for the 601.
Joe
Joe,
I've already be chastened by several others for the slip of my eye when reading
the Moto ad in MacWeek. It occurred to me at the time that it would be a major
achievement to develop a package that could dissipate 85 watts…but I didn't
give it much thought. Obviously, my old, tired eyes missed the decimal point in
the table. Why did they use such an ugly typeface, anyway?
Thanks,
-rich-
>>I've already be chastened by several others for the slip of my eye…<<
I saw that, but not before Navigator had sent the message!
Turn the lights up! <g>
Joe
Joe,
"Turn the lights up! <g>"
Even using a hologen light can't fix the lack of discrimination of tiny
artifacts in my old eyes. Unfortunately, I had to get a magnifying glass to see
the decimal in the ad. I think that the typeface Motorola used was just too
"cutesy" for the purpose. I'd have rather seen the specs printed in a nice
Helvetica or even Bookman face. At least even small type in those faces is
readable.
Thanks for the suggestion though,
-rich-
–>
The only technical challenge is board placement of the cooling tower for
optimal heat dissipation.
<–
Ever seen some of Digital's custom chips? (For example, their BI bus interface
chips) They actually have aluminum cooling towers coming out of them … some
of them have 2 or 3 separate disks on them.
Joe