January 27,1988

		GigaMOS Systems Software Product Report.


			Overview

	The purpose of this report is to set forth the status, options,
resource requirements and schedules of the Software effort of GigaMOS
Systems.  The unifying core of GigaMOS Systems is the provision of
hardware and software to support A.I. applications of the future.
(Although we expect that such hardware and software will prove useful
in many other situations as well of course.)  Our focus is on tagged
as opposed to conventional machines.  LISP, of course, is our primary
language, there may be some opportunity to come to terms with C as well.

			Background

	The primary hardware project of the company is the Falcon
project, sometimes also known as the K project.  The Falcon processor is
the only known processor designed after it became feasible to locate the
entire primary "core" memory on the processor board itself (as opposed
to communicating over a relatively narrow "bus").  Among known processor
designs, it is the only one to incorporate CALL Hardware, in which it
goes a step beyond the most advanced of the RISC machines.  In short,
these features and others give it a performance advantage of at least a
demonstrated factor of 3 in benchmarks as compared to the current market
leader for comparable machines (the TI Explorer II).  Since the E-II is
currently running with a clock almost twice as fast, it could be
reasonably argued that we have an architectural advantage of
approximately a factor of five.  (Note that TI's anticipated Mac II
board is only approximately one half as fast as the Explorer II).


		       Additional Background References

	Appendix 1 details the history of the AI computer market.  A
capsule summary is that circa 1980 tagged machines comming out of MIT
openned the market by introducing landmark improvements in performance,
ability to run large programs, and debugging enviroment.  In the last
couple of years, conventional machines (SUN, DEC, etc) have made headway
at the expense of tagged machines, mainly due to considerably lower
prices (as much as a factor of 3 for comparable hardware) and the
promise of several companies whose mission was to produce AI software
for conventional machines (LUCID, FRANZ).  Also, arbitrary limitations
that disqualified earlier conventional machines (notably address space
restrictions and lack of virtual memory) have been largely overcome,
although the software picture for conventional machines does remain
murky in key respects.  Recently, a relentless industry drive to agreed
upon standards has become apparent.  At the hardware level, white hot
competition has resulted in outstanding values for the purchasors of
industry standard boxes, especially in periphal equipment.  The
resulting market forces are operating to reduce the number of viable box
vendors to approximately 4 (IBM-PC (further divided into IBM PS/2 and
Old-PC, Apple, DEC, and possibly SUN).  Of these, the APPLE MAC-II has
the most steam at the moment, and is sort of threatening to take
over the AI workstation world.  IBM is hampered by the OS/2 screwup.
Also, they are apparently threatening to sue would-be users of their
Micro-Channel Bus, which apparently accounts for Symbolic's decision
to defer their 386 style board in favor of their MAC-II board.  OLD-PC
also has OS/2 problems and in addition some technical limitations.
The visible champion of OLD-PC is COMPAQ.  DEC is relatively secure
and happy in its world, but is not really paying attention to AI.
DEC's actual benchmark performace is surprisingly bad.  SUN is
hanging in suprisingly well.  Their old machines are expensive but
no doubt very profitable.  Their new machines appear seductive, but
turn out sufficently limited as to force customers to buy the old
expensive ones.  SUN has succeeded in setting industry "tone" with
its (supposed) commitment to open standards.   AT&T and Xerox
have been announced as licensees of its SPARC RISC machine and
AT&T has signed an agreement to buy 20% of SUN for $200 million.
(This last, by the way, is causing quite a stir in the UNIX world.)
SUN is openly rumoring 100MHz clock speeds for some future version
of their SUN-4 chip (no doubt several years away).

	We assume here some familiarity as to what a tagged machine is,
etc, and will not go into that here. (See appendix 2, ....) Briefly,
although tagged machines are in some sense "right" and have various
important features, their perceived advantages for the ultimate user
(who presumably is operating a fully debugged program, etc) have been
eroding, and this combined with high price (>$50K in most cases) have
resulted in the market trends noted above.  One primary remaining 
differentiator has been the presence of volatility based Garbage Collection
on tagged machines and not on conventional ones.  This too shows
some signs of partially eroding in that Franz is testing a "generation
level" garbage collector for conventional machines.  LUCID, too,
is reported to be nearing product introduction.  These techniques
reduce a pause that was several seconds and scaled with the total
data involved in the application to on the order of a second.
However, during that interval, LISP execution is completely blocked
while on a tagged LISP machine, LISP execution can be intersperced
with the actual garbage collection effort.  Also, on the conventional
machine, some price is paid in normal execution speed to perform
the required checking.

	There has also been some pressure on LISP itself as a language
for applications delivery.  Cleverly designed C systems can get quite a
bit of milage, appear very snappy, etc.  There have even been efforts to
mechanically translate LISP to C (Inference).  For high end applications,
however, there is some evidence the penduluum is swinging back.  The
LISP to C translator has been pretty much acknowledged as a failure even
though it operated successfully, and the VAX has been found pretty slow
and cluckey, etc.

	To reverse the trend away from tagged machines, however, it is
clear that the price gap for comparable machines must be drastically
reduced.  Furthermore, tagged machines must be available in a form to
take advantage of cost effective mass market periphals.  And still
further, and most importantly new bottom line functionality
differentiators must be introduced in market usable form.  The prime 
candidate for such a differentiator is the MOBY Address system, for which
GigaMOS Systems has been issued a notice of intention to grant patent
application by the U.S. Government.  However, formidable obstacles
must be overcome to achieve a successful market introduction of Moby
Address.



			Statement of Objectives

	Our primary initial objective is to provide the Falcon with a
competitive basic software suit and operating system.  We intend to 
accomplish this in short order by porting the existing LAMBDA software
suite.  However, we recognize that much else is required as well.



Moby Address

Current Status

Path Ahead  (central issues)
  Central core
  "Lambda Compatibility"
  "TI Compatibility"
  New Standards and Systems.
    CLOS
    New Window System

Supporting Issues
  Networking