[comp.research.japan] Trip Report, Fujitsu Laboratories

rick@cs.arizona.edu (Rick Schlichting) (02/03/91)

--------------------------------------------------------------------------
    Trip Report -- Fujitsu Laboratories  (March 29, 1990)

    Richard D. Schlichting   (rick@cs.arizona.edu)
    Associate Professor
    Department of Computer Science
    The University of Arizona
    Tucson, AZ  86721, USA 

    [This report represents the personal opinion of the author, who was on
     sabbatical in Japan from Dec. 1989 through July 1990.  The sabbatical
     was supported in part by grant INT-8910818 from the NSF U.S.-Japan
     Cooperative Science Program.]
---------------------------------------------------------------------------

    On March 29, I visited the Fujitsu laboratories in Kawasaki outside of 
Tokyo.  My official host for the visit was Dr. Takao Uehara, who is the 
manager of their software laboratory.   Sanya Uehara, who works in the 
software lab, was also heavily involved in my visit, and the one who did 
most of the arrangements. He graduated from Tokyo Institute of Technology 
about 10 years ago and has worked at Fujitsu ever since.  Also involved 
was Moto Adachi from the human interface research division who I had met 
in Tucson prior to leaving for Japan in December.  (See addresses below.)
    I left the office about 8:15 and arrived at the front gate of
Fujitsu around 9:00 without too many problems.  After getting a
visitors badge, I was pointed to the main building -- a gleaming
20-story building that is apparently about 2 years old.  At that point,
I was directed to a conference room on the second floor where most 
of our discussions were to take place.  I was allowed to venture up 
there on my own, which surprised me somewhat after the "escort" 
requirements that seem standard at many U.S. laboratories.
    At about 9:25, Sanya Uehara came in and introduced himself.  We 
chatted for awhile, and he gave me glossies describing Fujitsu, Fujitsu 
Labs (which is technically a wholly-owned subsidiary), etc.  At 9:30, the
offical part of the visit began as Takao Uehara arrived.  He gave me
a short description of the lab setup and then put on a videotape
describing some of the activities. 
    While the tape was on, other people started arriving, including
Moto Adachi, Mijio Aoyama from the business switching systems division,
and Yasunori Kimura from the Artificial Intelligence lab.  Probably a total 
of 8-10 in all.  After the tape and introductions, I gave my talk on
failure-handling mechanisms for the SR distributed programming language.  
    Afterwards, there were two presentations about projects going
on at Fujitsu, one by Aoyama on their multi-processor digital
switching systems and simulator, and another by Kimura on their
5th generation project to construct a parallel machine.  In the talk 
on switching system, Aoymama first laid out the history and general 
hardware of such systems, both at Fujitsu and elsewhere.  He then described 
in more detail their systems, including an overview of architecture of
the software system written in Chill.  Their most advanced model uses 
multiple call processing processors as a way to incrementally increase 
capacity and duplex hardware for reliability.  Finally, he described a 
special-purpose simulator for the system that they had constructed; it 
runs on multiple Suns (currently 2) in a local-area network.  Interestingly, 
it does not simulate failures, so the reliability aspects have not been
tested using this vehicle.
    Kimura then talked about the projects in the AI lab.  He first
described the KL family of languages that are being developed at
ICOT as part of the well-known "5th generation" project sponsored
by MITI. The languages are based on Guarded Horn Clauses (GHC) and 
have a resemblence to Prolog.  The family consists of 3 or so languages, 
from the base one to a "user version" that runs at the top level.  
I should note that his knowledge of this work was first hand, as he had 
been assigned to ICOT for 5 years as part of Fujitsu's contribution 
to the project.
    He then went on to describe more of the hardware architecture,
which is based on multiple processing elements organized in clusters.
They are currently in the process of designing the PE's, which
are intended to be optimized for KL.  They are apparently under
a great deal of pressure to get the entire machine built before
the end of the project, which is now only 2 years away.  In
addition to the usual difficulties, they also have the problem
that the language specs keep getting changed.  (When asked when
he thought that would stop, he said in 2 years when the project
ended!)  There are 5 companies in competition to develop the
hardware.  
    After eating lunch in a nice restaurant on the top floor
of the building, we spent a few minutes in an exhibition room
in the basement.  The bulk of it consisted of "hands-on" exhibits of a
lot of the current Fujitsu products.  In addition was my personal
favorite--a working model of a 1954 Facom computer built using
mechanical relays!  They cranked it up for me and after a lot of
clattering, it printed out 3 or 4 powers of 6.  
    After lunch, S. Uehara escorted me to another building to rejoin
Takao Uehara, who was to accompany me to three demos.  The building
in which they were located is apparently part of the software lab managed
by Uehara, and was quite a bit older than the main building.  The first 
demo was a parallel machine called CAP (Cellular Array Processor).  After 
describing the architecture, I was shown some sample graphical and heat 
simulation demos.  The existing machine is apparently the 2nd generation, 
of which 1 has been shipped.  They are currently starting work on the next
generation.
    The second demonstation was of a simulation of "neuro-computer." 
After a short description of the concepts, I should shown an application
consisting of a curve-fitting application using a 3-level neural network
on-line.  I was then shown a video tape that elaborated on the concepts, 
described out they are used to build "machine that can learn", and showed 
a cute robotic application.  One was dressed up as a robber, one as Sherlock 
Holmes, and two as policemen, with the object being for Holmes and the 
policemen to capture the robber.  The robber had been taught to flee, while
the others had been taught to follow the robber with varying degrees
of skill.  At the very least, it certainly was entertaining.
    The final demo was of a machine translation system (Japanese -->
English) that was recently released commercially.  According to
the person doing the demo, it's being used to do such things
as translate handbooks and manuals, applications where some of the
resulting awkwardness can be tolerated.   He described the basic structure 
of the system, which uses an intermediate form called "interlingua", and 
then ran some sample sentences.  The samples were intended to show
how the system can make some fairly fine distinctions, and
I actually found it pretty impressive despite the fact that the
person doing the demonstration downplayed its sophistication.
The algorithms were developed by the people in the lab, who then gave 
it to the products division who did things like add an editor front-end 
to it.  They are apparently working on a version that will translate 
English to Japanese based on the same interlingua idea.
     That ended the official part of my visit.  I then sat with
Moto Adachi for an hour or so chatting about various things.  Among
the more interesting were his comments on how it was getting hard
to hire new people because starting salaries are low compared to 
places like banks and insurance companies.  This aspect of the Japanese
labor shortage has been given much attention in the English-language
press as well.  I found this to be an interesting difference from the
situation in the U.S., where my perception is that it would be
unusual for a person with a technical degree -- especially a graduate
degree -- to consider going that route.

Addresses
---------
Dr. Takao Uehara 
Deputy General Manager, Information Processing Division
Manager, Software Laboratory
Fujitsu Laboratories, Ltd.
1015 Kaminodanaka, Nakahara-ku
Kawasaki 211, Japan

Sanya Uehara 
Software Laboratory
Fujitsu Laboratories, Ltd.
1015 Kaminodanaka, Nakahara-ku
Kawasaki 211, Japan

Motomitsu Adachi
Senior Engineer, Human Interface Research Division
Fujitsu Laboratories, Ltd.
1015 Kaminodanaka, Nakahara-ku
Kawasaki 211, Japan

Mikio Aoyama
Senior Engineer, Business Switching Software Dept.
Business Switching Systems Division
Fujitsu, Ltd.
Fujitsu Nakahara Bldg.
629 Shimokodanaka, Nakahara-ku
Kawasaki 211, Japan

Yasunori Kimura
Artificial Intelligence Laboratory
Fujitsu Laboratories, Ltd.
1015 Kaminodanaka, Nakahara-ku
Kawasaki 211, Japan