United States. Congress. Senate. Committee on the.

The Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) online

. (page 58 of 140)
Online LibraryUnited States. Congress. Senate. Committee on theThe Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) → online text (page 58 of 140)
Font size
QR-code for this ebook


the companv was only moderately profitable.

Back in 1959, Dr. Bernard Rothlein left Sperry (Rand) Semiconductor with
a small group "and formed National Seminconductor. Sperry sued witli a ven-
gence and in the court room succeeded in dramatically winning its case. Its
trial attorney used the Sperry Semiconductor organization chart as his primary
exhibit and proceeded to place large black squares over the face of each defactor.
Sperrv won a sizeable judgment against National, nearly forcing the company
out of business. It managed to keep going, however, and in 1967 Peter Sprague,
a wealthy voung investor whose father founded Sprague Electric, bought control
of Natioiial and persuaded Charles Sporck. then general manager of Fairchild.
to join National as president. Sporck took with him four key men. Fred Bialek,
Pierre Lamond. Roger SmuUen. and Don Valentine, and National got off to a
roaring start. Bob Widlar. the wizard designer of Fairchild's most successful
circuit, the linear oi^erational amplifier, the 701. was already at National. Widlar
redesigned the 701 and National's 101 circuit became the industry standard.

The most successful spin out from Fairchild is Intel, an acronym for Integrated
Electronics. Dr. Bob Noyce and Dr. Gordon Moore, two of the original eight
Fairchild founders, left, along with Dr. Andrew Grove, to pursue the MOS semi-
conductor memory market which they believed could shortly become price
competitive with 'ferrite cores. Intel was founded in 1968 with backing from
Arthur Rock, the original backer of Max Palevsky of Scientific Data Systems.
Five years later, in 1973. Intel had sales of $67 million and had successfully
perfected the processing techniques necessary to make high density MOS chips
at low cost.

Obviou.sly Fairchild had to replace Dr. Noyce, and so it recruited Dr. lister
Hogan froln Motorola, who at the time was preparing to take a group with
him and join General Instrument, an early entrant in the MOS business which
even today is relatively unsuccessful. Sherman Fairchild was determined to

40-927 O - pt. 7-33



5422

hire Dr. Hogan and his offer reportedly included an interest free loan to purchase
$1.0 million of Fairchild stock under options. Dr. Hogan brought with him eight
others from Motorola, who were later to become known as '"Hogan's Heroes."
This move only increased the number of spin outs from Fairchild as veterans
were replaced with their Motorola counterparts. Within a year there were four
additional spin outs, the most successful of which was Advanced Micro-Devices,
a firm devoted to second sourcing bipolar devices.

The semiconductor industry debacle of 1970 dried up the stream of venture
capital which had supported the startups, and there have not been any new-
companies formed in the past three years. National recently lost one of its
founders to another segment of the industry, and it is iwssible that a spin
out might come from Intel in the future, but it appears as though after nearly
two decades the formative and confusing years are at last over. Xo mention
has been made of the two largest companies in the industry and for good reason.
Texas Instruments has spawned only a few new companies, the largest of which
is Mostek with backing from Sprague Electric ; IBM Components Division has
spawned only two companies, Cogar on the East Coast, which went bankrupt
about two years later, and Advanced Memory Systems, which is currently having
rather severe problems. A third spin out. Semiconductor Electronic Memories,
was financed by Electronic Memories and Magnetics about four years ago, but
has yet to develop significant revenues. The schematic Figure 5 shows the
genealogy of the semiconductor industry since its inception in 1947.

IBM set up its first West Coast facility in 1943 which was a punch-card
plant, but it was not until 1952 that the company established a research center
in San Jose, California. It was at that plant that the magnetic data storage
disk was invented in the 1950's and that development resulted in a proliferation
of spin outs to pursue the new storage medium. It is also true that IBM veterans
have formed companies engaged in many other aspects of computer technology,
predominantly in the peripheral equipment area. However, only three groups
have left the IBM Components Division to strike out on their own and none
of them have been successful. The reason, as the following discussion will show,
is that IBM's semiconductor exix'rtise lagged the industry until quite recently.

Texas Instruments was one of the twelve original licensees of Bell Telephone
Laboratories in 1952 and was in high volume production of germanium tran-
sistors in the mid-1950's. In 1954, TI announced the first silicon transistor which
w^as available for certain military programs in 1956. Fairchild announced the
double diffusion planar process for silicon transistors in the late 50's and was
in volume production in 1961. Back at TI, Jack Kilby invented the integrated
circuit in 1958 and the first IC's were commercially available in 1964. IBM's
first computer, the 701, used vacuum tubes and that machine was produced until
1959. The 1401 was introduced in 1959 and used germanium transistors, although
silicon transistors had been available since 1956. It is difficult to pinix)int the
exact time when IBM decided to go into the components business, but in 1954
research and development expenditures leaped to 7 to 8 percent of sales from
3 percent the previous year. Since the first computer (the 701 had been intro-
duced in 1953 and IBM w^as already planning its next machine, the 1401)
we can only assume that management set up and R&D facility in 1954 to pursue
the still relatively new transistor technology which was already in volume
production at TI. All the vacuum tubes for the 701 computer were purchased
from outside suppliers, and the introduction of the transistorized 1401 in 1959
caused a slump in the vacuum tube industry as 701 production was phased out.

The early silicon transistors were quite costly and the very high reliability
devices were used exclusively in military programs. Those devices which did not
quite meet military specifications were sampled by various commercial enter-
prises so that their performance characteristics could lie designed into equipment
which was then on the drawing boards. IBM, however, elected to remain with
geranium transistors, which it was then producing in-house, and these devices
were used in all the 1401's which were phased out of production in 1964. IBM's
largest outside suppliers were TI and Fairchild, but they were used as second
sources to fill gaps in IBM's internal supply lines.

By 1960-61, IBM was in the early design stages of its 360 computer, which was
to be a family of machines with common architecture, software, and peripherals.
The heart of this machine was the components. In spite of the fact that the inte-
grated circuit had been invented at least four years earlier and could easily
have been in volume production at IBM and TI and probably Fairchild in 1963,
IBM made the decision to use a hybrid circuit which it called Solid Logic



5423

Technology. The decision to use SLT v.as made in 1961, but the first 360 was
not introduced until 1964. SLT was an early form of transistor-transistor-logic
(TTL) and was ultimately extremely easy to produce. Each package was built
up from a one inch square of ceramic substrate to which the various components
were added — the resistors, diodes, capacitors, and transistors. The technique
lent itself to extensive automation, and therefore had a very low labor content.
Moreover, it was very cheap because the package could be reworked if it didn't
pass quality control tests, which gave the packages fantastic overall yields.
However, these packages were not availal)le outside IBM and production was
absolutely critical to the success of the 360 computer — it was a "you bet your
company" product. IBM's organization charts show just how critical these com-
ponents were ; the Components Division operated semi-autonomously until mid-
1963 when the entire division was suddenly pulled under the jurisdiction of
Vincent Learson, where it remained until the end of 1966 when shipments of
the 360's were back on schedule, presumably because SLT yields were finally
satisfactory. IBM stayed with the SLT technologj- until the introduction of the
370's in 1970. although the independent semiconductor companies had by then
introduced many other faster and denser devices. Was IBM cautious, or was it
deliberately bringing the heart of its machines in-house? We think the evidence
shows that the latter answer is correct. If IBM was being cautious, it could
have selected several of the semiconductor houses which were proliferating in
the early 60"s as suppliers, thereby guaranteeing itself at least one second source.
But it was only in 1966-67, when the Components Division was dismissed from
Learson's jurisdiction, that the decision was made to phase out of the manu-
facture of all discretes and purchase these devices from outside suppliers. IBM
also purchases some other devices on a bid basis, but none of the suppliers really
know where the circuits are being used. As one supplier phrased it, "IBM comes
in with some specs and wants a bid. We don't know where the circuit is going
and we don't know all the parameters, but we all bid. It's a real horse race."
We estimate that IBM purchases about $50 million of devices outside and
produces about $600 million.

When IBM introduced the first 370 machine in 1970, the circuitry used was
another in-house design called Monolithic Systems Technology (MST). How-
ever, in 1965 Texas Instruments introduced the most successful logic family in
the industry, transistor-transistor-logic (TTL), in the form of the 5400 for mili-
tary applications and shortly thereafter, the 7400 for commercial applications.
Initially, TTL was produced in small scale integration ( SSI ) only, but by 1969
TI and a few others were building medium scale integration (MSI), and by the
time the 370 was introduced large scale integration (LSI) was readily available.
Also. Motorola had introduced an even faster device known as the MECL-10,000
Emitter-Coupler-Logic for very high speed computer applications. These devices
are used very successfully by all the other computer companies and are produced
by at least one dozen suppliers. Yet IBM again decided to do its own thing and
produce internally, not necessarily with a high degree of success.

The history of the devices used by IBM in .the 370's is revealing ; in June 1970,
the 155 and 165 were announced with core memory. In March 1971, the 135 and
145 were announced with bipolar memory, but in July 1971 the 195 was introduced
with core memory. In August 1972, IBM announced new versions of the 155 and
165 (which had used core memory), the upgraded 158 and 168 which used MOS
memory. The last two of the 370 machines to be introduced, the 125 in October
1972 and the 115 in June 1973, both use MOS memory. Again one might ask if
IBM were merely being cautious and the answer has to be negative. In 1970
when the first 370 was announced using core memory, several suppliers were
already building MOS memories in small sizes, 256-bit devices, and Intel's 1024-
bit device, the industry standard #1103 was available. Bipolar devices were also
available on a limited l)asis and volume production was in place in 1972. More-
over, in the 1970-71 period the semiconductor industry was in the throes of a
real depression directly related to IBM's introduction of the 370 system as u.sers
cancelled orders from other computer manufacturers, and as IBM's outside pur-
chases of discrete devices for the 360 series .stopped. The MOS memory houses
meanwhile were attempting to convince the otlier compiiter and peripheral
houses that this technology was the future trend. Data General was considering
purchasing MOS memories from Intel in 1971, but IBM had just introduced the
135 and 145 with bipolar memory, and little Data General uith $50 million in
sales could not risk an untried technology so it stayed with core. Partly as a re-
sult of the dislocations created by termination of 360 production and lower
purcha.ses of semiconductors by the military, the industry did not add capacity
in 1970 and 1971 and was unprepared for the 1972-73 explosion in demand.



5424

But in August 1972, IBM announced the upgraded versions of the 155 and 165,
the 158 and 168, using MOS memories. IBM had finally blessed the technology
and there was a wild scramble for MOS parts which the semiconductor industry
was unable to produce in sufficient quantities ; it is only very recently that sup-
ply has finally caught up with demand. Meanwhile, the early purchasers of the
155 and 165 with core memories, and there are an estimated 2.000 installations
of these machines with monthly rentals of $50,000-$100,000, were offered the op-
portunity to upgrade their new but obsolete machines at a cost of about $300,000.
And so it was that IBM, the industry leader, simply used the oldest technology
available, core memory, until it was able to produce its internal needs which took
it two years. The history of semiconductor industry introductions and IBM's
usage of electronic circuitry is shown in Table III.

Now that IBM has MOS technology perfected, it is apparently gaining a tech-
nological lead over the semiconductor industry for the first time. The 135 and 145
machines each use bipolar memory and IBM is putting four 1024-bit (4— IK)
chips on a ceramic substrate to form a 4096-bit (4K) module. We know from
papers delivered to professional meetings that IBM has the capability of build-
ing a single 4-K MOS device and that it has had some limited success with a 32-K
MOS device on a single chip. MOS is inherently cheaper than either bipolar or
core memories although it is slower, but because the densities can be signifi-
cantly higher, most memory circuits over the next 3^ years will probably be
MOS. What IBM is apparently planning to do with the 4— K bipolar module in
the 135 and 145 machines is to replace that module with a single 4— K MOS chip.
It can do this by using a buffer memory which can look back into any kind of
memory and ahead at any kind of logic. Therefore, as IBM gets down the learn-
ing curve, which it can do more quickly than anyone else because of the huge
volumes of only a few kinds of devices, it can simply replace existing modules
with single chips ; for instance, it is theoretically possible to replace four 4-K
MOS chips with a single 16-K chip, or two 16-K chips with one 32-K chip. The
other computer companies will not have this capability for a significant period of
time because their suppliers, the semiconductor companies, must make a profit
on the components which they produce and they cannot produce these chips
profitably for some time.

In the semiconductor process, the most critical factor in profitability is yields,
the number of good chips from a wafer, and the most critical factor in good
yields is good chip design. At the present time the semiconductor industry is
struggling to produce 4— K MOS parts in quantity, but is unable to do so because
of very low yields, which in turn make the parts too expensive to be cost com-
petitive. Each three inch silicon wafer, which is the building block for the chips,
can contain 200 potentially good chips, each 21,000 square mils (the smaller the
chip size, the more chips on a wafer ; the better the yields, the higher the profits) .
In the early to middle stages of production, the yields are terrible ; initially
there might be one good chip out of every other wafer, then you would get one
good chip from every wafer, then perhaps two good chips per wafer, and so on.
In the middle stages of production, a 10 percent yield would be considered abso-
lutely fantastic ; that would yield twenty devices at the wafer fabrication level.
However, the semiconductor process is terribly complex and these die must then
go through additional manufacturing steps which include scribe and break, first
optical, die attach, bond mold or seal, and final test. At each level in the process
more die are lost, so that the final yield is usually 50 i^ercent less than at the
wafer fabrication level, or in this case ten parts per wafer. At present, the semi-
conductor industry is still in the early stages of producing the 4-K chips so
that yields are still at very low levels, perhaps five good chips per wafer. By
1976, we think yields could triple to about fifteen parts per wafer, which at an
average selling price of $14.00 in OEM quantities, would produce a profit of $100
per wafer, the absolute minimum required to cover the other costs of doing
business.

IBM does not have the same cost reqtiirements as do the semiconductor com-
panies because its profits are derived at the system level and components do not
necessarily have to be profitable items when they are first used. On the other
hand, the semiconductor companies cannot afford to sell cliips to the computer
mainframe companies at a loss even though these chips might make their prod-
ucts more cost effective. The IBM Grey Book on the 135. which was one of the
exhibits in the Telex trial, details IBM's strategy in the components area and
emphasizes the importance of the new technologies as they relate to IBM's costs
over the life of the product. Exhibit Xo. 115, page 43 "* * * The Memory tech-
nology. Phase 21. absorbs about 50 percent of each new build cost, while the



5425

MST-2 logic cost accounts for another 25 percent. Both technologies are cost
sensitive to quantity variations and yi'ld percentages. The performances of East
Fishkill. Burlington, and Endicott CPM are, therefore, as vital to the financial
success of tlie M135 as the effectiveness of Kingston in providing the power and
the assembly and test functions * * *." In other sections of this same docu-
ment IBM details its strategy in pricing the CPU and the memory, again de-
pendent upon yields and upgrading Phase 21 to the newer technique. If, as IBM
seems to suggest, 75 percent of new build cost is accounted for by components,
then the other computer mainframers are indeed at a great disadvantage, but
the reasons are far more complex than just the cost of components.

IBM is quite probably the most profitable semiconductor manufacturer on an
unallocated cost basis because it does not produce a broad line of circuits, only
tho.sie which it uses in its own products. Volume is the key to profits in the com-
ponents business, and IBM clearly has the longest production runs of any of the
houses. What other house can run the same product with no iterations seven days
a week for two or three years? IBM's research and development activities in the
components area are also extremely effective because it can allocate all of its
expenditures to the improvement of technology for computing purposes. It does
not need to open new markets in the consumer area, such as hand held calculators
or electronic watches to assure future growth, nor does it have to market to
several different classes of customer. In 1973 IBM .spent $730 million on research
and development : if we as.sume that at least 10 percent of that was spent in the
components area, then .$73 million was spent on improving components for com-
puter usage. The semiconductor industry spent an estimated $150 million on
research and development in its four major markets, government, consumer, in-
du.strial, and computers. If those dollars were allocated equally, this would mean
that 25 percent of the $150 million was spent on computing uses, or about $38
million, approximately half the amount spent by IBM. In reality, we suspect
that IBM spent more than $73 million on components since we have shown how^
critical they are to the computer business, and would not be surprised if that
figure was closer to $100 million, or three times the amount spent by the other
semiconductor manufacturers for improving computer products.

These figures become even more distorted when one considers that the $38
million spent by the semiconductor industry w^as spread among five computer
mainframe companies, about a dozen minicomputer manufacturers, and at least
100 other computer-related customers. Each of these customers has to develop
a close w^orking relationship with its components supplier because the needs of
each are different from the other. Therefore, creative manpower is also spread
among the various customers, but more importantly, the supplier must know
the precise needs of the customer in order to offer the optimum circuitry. This
was not true when computers used transistors but in today's technology an out-
side supplier effectively has control over system performance instead of the
designer of the system maintaining that control — except IBM. When IBM
begins the design process on a new computer family, all of its needs are avail-
able in-hou.se, especially the components division. Questions are po.sed and
immediately answered — yes we will have a 4-K chip ready in 1974 if we are
given X dollars. The profitability tradeoffs can then be figured, and if it makes
economic sense to use four 1-K devices until Components is altle to make a
4-K device, it really doesn't matter becau.se the turnaround time among all di-
visions can be integrated so that no slowdown in production occurs. Perhaps
more important than anything else is the absolute secrecy which is maintained
internally because no outsiders need to be involved.

As the above discussion shows, IBM is moving from a follower to a leader
in semiconductor technology and the pace appears to be accelerating, so that
IBM will clearly be at the forefront in semiconductors for computing usage
in a few years. When Control Data sued IB^NI, part of the relief it requeste<l
was that IBM's Components Division be spun out so that the other computer
mainframers had the benefits of what was then only beginning to be technical
superiority in the semiconductor area. IBM admitted de facto that its Com-
ponents Division was critical by dismantling it as a separate unit in 1971 and
.spreading it over all other divisions so that it was not easily .separable.
IBM's semiconductor expertise must be made available to the other computer
companies in order to assure future competition. Furthermore, the Semiconductor
Test Equipment Division must be spun out as well since it is a necessary adjunct
to the use of made-in-IBM components.



5426

PART V. THE SEMICONDUTOE INDUSTRY WITH IBM AS A COMPETITOR

As we have pointed out, IBM has recently gained a technological lead in
semiconductors as a result of devoting $70-$100 million to research and de-
velopment for computing use, compared with an estimated $38 million by the
semiconductor manufacturers. Table III on page 37 shows the lead which IBM
is believed to have in producing 4-K MOS memories and the expected lead it
will have over the next several years, which in effect is a quantum jump. How-
ever, another factor of great importance is that this quantum jump could very
possibly involve a completely new type of technology which IBM will have
perfected that will not be available industry-wide until the 1980's. At present,
IBM is known to be conducting research on charge coupled devices (CCD's)
and the other semiconductor houses are working on this type of circuitry also.
In fact, a few CCD circuits have been announced by Fairchild, although they
are not yet available in any quantity, if at all. The CCD technology is akin
to MOS processing techniques and I have no doubt that the semiconductor
houses will be able to produce them in time. However, there is some question
as to when the semiconductor houses will be able to produce 8-K or 16-K
MOS memories because the present photolithography method is marginal at that
degree of density. It is probable that electron beam lithography will be required
to produce the high degree of resolution necessary to properly diffuse the wafers,
and this equipment is not available as yet. This creates a real problem for the
other computer mainframers in the 1976-77 time frame because IBM is certain
to use its technological advantages in designing the FS. expected to be intro-
duced about that time.

Further on down the road, IBM may elect to use bubble memories which are
totally unlike semiconductors. Bubble memories are not produced from silicon,
which is the basic material used in all other semiconductors, including MOS.
CCD's do use silicon but they represent an advancement of about six years in



Online LibraryUnited States. Congress. Senate. Committee on theThe Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) → online text (page 58 of 140)