home *** CD-ROM | disk | FTP | other *** search
- From: blalock@hpl-otis.hpl.hp.com (Travis Blalock)
- Date: Tue, 17 Nov 1992 20:31:52 GMT
- Subject: Re: Re: Lifetime of 25MHz IIsi (was: 25MHz IIsi and my engineering friend)
- Message-ID: <71610001@hpl-otis.hpl.hp.com>
- Organization: HP Labs, High Speed Electronics Dept., Palo Alto, CA
- Path: sparky!uunet!pageworks.com!world!eff!sol.ctr.columbia.edu!usc!elroy.jpl.nasa.gov!sdd.hp.com!hpscit.sc.hp.com!scd.hp.com!hpscdm!hplextra!hpl-opus!hpl-otis!blalock
- Newsgroups: comp.sys.mac.hardware
- References: <1992Nov16.181708.22881@news.uni-stuttgart.de>
- Lines: 81
-
- In comp.sys.mac.hardware, isr@rodan.acs.syr.edu (Michael S. Schechter - ISR group account) writes:
-
- > they'll be more voltrage drop, and the CPU will draw still more amperage,
-
- Ack! Less voltage --> lower current
-
- Actually, in the CMOS case, the output capacitances have to be charged to
- lower voltages when the supply voltage decreases. As a consequence the
- average current goes down. Unfortunately, the max speed of the logic gate
- goes down since the current available to charge output capacitances is
- quadratically (to a first order) related to the supply voltage. This is why
- low voltage circuits are so difficult to design for high speed.
-
- As regards to the 25Mhz upgrade of the IIsi: Power dissipation in CMOS
- is dominated by the currents necessary to charge and discharge gate load
- capacitances. This form of power dissipation is given approximately by:
-
- P = C(load)*[V(supply)^2]*frequency
-
- Therefore, for lower V(supply) and constant C and f, power dissipation
- goes down (quadratically). Since power is a linear function of frequency,
- 25% increase in clock frequency generates a 25% increase in power
- dissipation.
-
- In terms of reliability, heat is the enemy of chip reliability. It was
- mentioned earlier that diffusion of junctions would lead to failures but
- that is not an issue at these temperatures. Electromigration is
- accelerated by heat and will most likely cause the failure. For the
- curious, electromigration is a slow migration of the metal conductors on
- a chip due to a combination of heat and high current levels. This
- phenomenon can lead to open circuits. In any case, adding a good sized
- heat sink to the processor chip will improve the expected reliability
- beyond that of an unmodified IIsi (if the chip runs at a lower
- temperature than it did before). Not adding the heat sink will reduce
- the reliability. The question is whether the reliability reduction is
- noticeable, i.e. Does the MTBF go from 30 years to 20 years, or from 10
- years to 2 years? If it is reduced to 20 years, then you will be using
- your wizbang third generation PowerPC with the neural interface by then
- so who cares. Since I don't know the answer, I would add the heat sink
- to be conservative.
-
- Another point in favor of the heat sink is the improved performance of
- CMOS at lower temperatures. For a given input drive voltage, the output
- current of a MOS transistor decreases with increasing temperature. This
- is caused by the reduction in the carrier mobility as temperature
- increases. Again for the curious, lattice vibrations in the
- semiconductor increase with increasing temperature so the current
- carriers (electrons in an NMOS device) have a reduced 'mean free path'
- i.e. they bang into things more often and take longer to get through the
- device. Could you get through a crowd of people slam-dancing faster
- than a crowd of people standing perfectly still? Probably not. Of
- course, this is a simplified explanation but you get the idea. Since
- chips are spec'd to run over a range of ambient air temperatures
- (usually up to 85C), it stands to reason that a chip that will run at
- 20Mhz at 85C should run faster at room temperature (25C). If 25Mhz is
- at the outer edge of the chip speed margin, then running without a heat
- sink can raise the temperature enough to degrade the performance so that
- the chip will begin to have failures (i.e. dropped bits...) as it heats
- up. Again, your chances of success are improved if the temperature of
- the chip is held lower.
-
- All of this is focused on the main processor since we are working on the
- assumption that the rest of the ASICs were designed originally for the
- IIci and should be fine for 25Mhz operation. This assumption seems to
- be at least partially correct since most of the modifications are
- working.
-
- Bottom line: Increasing your clock speed could work fine for several
- weeks but you might crash in the middle of a big edit during a hot day
- next August when the AC in your room isn't working so well. Or you
- might crash due to an init conflict tomorrow (hey, this is a Mac :-)).
- As always, save early and save often. Try it if you feel like it but be
- aware of the risks.
-
- I hope someone found some of this useful.
-
- Travis Blalock
- blalock@hpl.hp.com
-
- p.s. My 25Mhz IIsi is working fine so far!
-
-