Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!uxc!uxc.cso.uiuc.edu!mcdurb!aglew From: aglew@mcdurb.Urbana.Gould.COM Newsgroups: comp.arch Subject: Re: quest for breakthroughs (long) Message-ID: <28200278@mcdurb> Date: 18 Feb 89 00:41:00 GMT References: <740@tetons.UUCP> Lines: 232 Nf-ID: #R:tetons.UUCP:740:mcdurb:28200278:000:10128 Nf-From: mcdurb.Urbana.Gould.COM!aglew Feb 17 18:41:00 1989 >/* ---------- "quest for breakthroughs (long)" ---------- */ > > This news group is stuck in an endian rut and strung out on history. >How about an exercise in creativity? The goal is to use computer >architecture as the basis for directing future technological advances. > > Imagine that you are the enlightened head of Imaginary Computer Corp's >architecture department. Your job is to tell us (the enlightened >scientists at the research lab) exactly what sort of technological >breakthrough would help you the most. OK, I'll bite. Actually, I'll probably bite several times, with different sets of starting assumptions/markets, as time permits in the next few weeks. First, I'll think about my dream, of building a truly useful personal computer system... > What are your assumptions? >End Product: Embedded controller, PC, Workstation, Mini, SuperMini, > MiniSuper, Mainframe, Super, ... PC: Subclasses - desktop (where most PCs are now), briefcase, wristwatch, house. >Architecture:RISC, CISC, Vector, Massively parallel, VLIW, Shared memory > multiprocessor, ... Whatever it takes. Probably RISC/CISC single processor. Actually, I don't think that processor technology is all that important for this market, although performance increases in a single processor may help in getting cost down by eliminating extra components. >Application: Home, Business, Engineering, Scientific, Manufacturing, ... Home, Engineering, Scientific (basically, the personal computer I want) >Timeframe: Next year, in 5 years, in 10 years, ... 5-10 years. > What problems are you trying to solve? >- Performance, Cost, Complexity, Size, Reliability, ... I'm sure I'm going to get told "You want a personal scientific computer? What about NeXT?" My response is that NeXT is much too expensive. I want a really cheap computer, <2,000$, preferably small enough to wear on my wrist, with ability to interface to the standard I/O modules. A 33MHz 68030 based system with 32 MB of memory is the first small computer system that I've found acceptable to work on, so I don't think processor performance improvements are too much needed for my "realizable" dream machine (although it still falls a way short of a Gould NP1 with 256 MB, that's chiefly I/O). So, here's what I want to solve: Performance comparable to today's top of the line workstations with considerably reduced component count. With a view to this, performance sufficient in a single processor to eliminate the need for separate graphics coprocessors, etc. (or, multiple processors per package - chip or hybrid). Cost: get it *way* down! Complexity: reduce component count to reduce cost Size: Second priority - make small enough to fit on briefcase, wrist... Reliability: Third priority - reliability comparable to one of today's PCs would be acceptable (but cost of repair should also stay same, which it probably won't - it'll increasingly be "Buy the whole thing") > Breakthrough examples: >- Practical X-Ray lithography creating 1 million gate CMOS chips Anything that increases densities is good. >- Quantum transistors and high temperature superconductor "metallization" Probably not applicable to this domain. I'm not going to carry liquid nitrogen around. > layers creating femtosecond propagation delays Probably not too applicable to the performance domain I'm talking about in this note (will be to others) >- Fiberoptic advances creating 1 Gigabyte/sec cable or bus bandwidths High priority. What I want is a wristwatch computer that I can plug into a base unit to upload/download stuff from disk, control display devices with, etc. (Or, I want a base unit PC that can download stuff to a wristwatch unit - but I would really prefer that the wristwatch, or Walkman, unit have the intelligence, with enough non-volatile memory to be "THE" system, with everything bigger a peripheral). Yes, I'm a portable computer freak. I bought one of the first luggables, and lugged it thousands of miles. At the moment I don't have a portable; I do have one of those multifunction memory watches. But, these watches are sadly limited, with far too little memory and I/O capability for what I need. I would have bought one of those SEIKO memory systems that had a PC interface, except that I didn't have a PC compatible. I think that they missed the market with that device - instead of having a simplified ASCII download capability, perhaps with a stripped down RS232 i/f (I think that you could get that down to wristwatch profile) they went for PC users. PC users don't want writswatch computing. The people who want wristwatch computing are the people like me who use bigger machines all day long, and want the ability to manipulate their wristwatch schedule, appointment books, etc., with considerably more convenience than a writwatch keyboard allows (you know how much pain it is to type things in on such a keyboard? Give me download!) I want a personal computer that can replace the notebook I carry around all day, that can interface to other devices and computers. TABLET is close, but I don't want to have to carry that around. >- Optical disc advances creating terabyte 3.5" optical discs Yes! Appears possible. Myself, I would just wait for that to come out for the desktop crowd; I can't see auxiliary storage being writwatch sized in the near future. Walkman sized maybe. In my own company I would concentrate on the wristwatch processor memory and I/O element that nobody seems to be working on. >- Nanotechnology inventions creating microscopic computers Sure. But I don't see this being rewarding in the timespan I'm talking about. > Keep in mind system constraints like memory and IO bandwidth, power >and cooling, packaging limitations, reasonable economic assumptions, >and balanced system design. For instance an engineering workstation >manufacturer may dream of femtosecond gates in 10 years, but the >development and manufacturing costs would probably make the technology >prohibitively expensive for a workstation in that timeframe. The memory >and IO to support that kind of cycle time also would not fit with >workstation cost, size, and software. What I want is basically a present day processor, 8-32MB of memory, and an "optical SCSI" in a 1"x0.25"x0.10" package. Everything else in this "realizable dream" I see on the drawing boards. What's needed: is increased density; technology that permits logic, memory, and optics to be on the same chip; a good way of coupling optics to a chip (that doesn't require a relatively large shroud); packaging to protect this beast; and manufacturing technology to make this beast buildable (I can't see it being at all practical without robotics or *very* cheap labour); and, of course, reductions in power consumption so that I could wear such a device, or attach it to my belt. The worst part about this dream - although I think it's within reach, I don't think that it would be a big profit maker. After all, the idea would be to make something *cheap*. It'll probably arrive 5-10 years after I would like it, via excess competition in the PC market forcing people into new niches, or (more likely) via intelligent bank cards. > Try not to expect too many breakthroughs at once, magically >eliminating bothersome constraints, but also making the scenario >implausible. For instance, femtosecond propagation delay gates are >highly unlikely in commercial products in the next 10 years. First, >they will require breakthroughs in the commercialization of quantum >transistors. Second, in order to derive real benefit from them, >breakthroughs in reducing on-chip and off-chip wire delays will also >be necessary. The combination of two breakthroughs at the same time >strains credibility (but isn't impossible.) I'm trying to be mid-range practical. Please tell me where I fall down (economics is the biggest constraint, I suspect) > At what point do technology changes affect the architecture? At what >point do you get diminishing returns? > > Chip density: 50K gates -> 100K -> 500K -> 1M -> 5M -> ? Say 24K gates for the processor, + 8M (minimum) of RAM. O(16M) gates necessary if all on same chip; less would require great pinout. > Chip pinout: 250 pins -> 400 -> 800 -> 1000 -> 2000 -> ? If memory and processor could live on same chip, both space, heat, and process-wise, then low pinouts need. Direct coupling to 1 or 2 optic fibers on same chio desirable. > Propagation delays: 500ps -> 100ps -> 20ps -> 5ps -> 500fs -> ? Not applicable. > Chip power dissipation: 1 Watt/chip -> 10 -> 20 -> 50 -> 100 -> ? I don't want to wear anything that hot on my wrist! How about fewer watts per device! (which is probably where I start needing new technology - I suppose that there are physical limits here) > Cable bandwidth: 100 Mbits/sec -> 500M -> 1G -> 10G -> 100G -> ? The basic thing I need for this wrist configuration is the ability to do bitmapped video, say for a colour megapixel at 60 Hz => 24x1Mx60 => O(2G) bits per second. Plus the rest of I/O traffic, which is relatively small. Biggest need is miniaturization of terminators. > > The intent of this exercise is to discuss what kind of technology advances >really benefit a particular computer architecture. You may want to attack >the problem from the other side, starting with a breakthrough and >determining which architecture would be most suitable. > > >Apply: Standard disclaimers >-- > Bob Blau Amdahl Corporation 143 N. 2 E., Rexburg, Idaho 83440 > UUCP:{ames,decwrl,sun,uunet}!amdahl!tetons!bb (208) 356-8915 > INTERNET: bb@tetons.idaho.amdahl.com This was a good idea. Writing this has been fun, and clarified a few ideas in my head. If I get the time I may try to write similar things for more realistic / commercially viable systems.