|Series 9||Integrated Circuit Engineering Collection|
A newsletter type report published by Integrated Circuit Engineering a minimum of 12 times a year.
from: GaAs - Still a Promise February 12, 1982.
For a number of years gallium arsenide has held the promise as a new, emerging semiconductor technology which offers considerable performance advantages over mainstream silicon technology. Every year, usually at the annual GaAs Symposium, the perennial champions stand and say, "the second wave beyond silicon technology is almost here."
Hardly a week goes by without a major systems house contacting ICE concerning technology assessment - with particular interest in GaAs. Will GaAs technology be required in their repertoire of technologies in order to compete in the marketplaces of digital data processing and telecommunications? The question has ominous implications; a $40M up-front investment in bricks, mortar, leasehold improvements, equipment and at least a $5M yearly budget for research, development and continued upgrading of the facility. Of course, this is in addition to the problem of staffing with knowledgeable people.
The shear magnitude of silicon technology dwarfs the investment in GaS - to a few percent of that invested in silicon. There has not been a wide appeal for a merchant semiconductor manufacturer to concentrate on one standard GaAs part or family of parts to underwrite the extremely high investment necessary to develop a widely-used component. The merchant market companies such as Intel, AMD, National, don't feel threatened and have not yet picked up the gauntlet. TI could possibly be the exception, with some research in GaAs for military applications. Note that the majority of recently-published papers and news releases* in this field are from Japanese companies which tend to supply materials or components for the merchant market as well as for their own computer factories. They certainly
are not engaged in this research at the behest of the Japanese Government for a domestic military market.
GaAs "gurus," such as Howard Phillips and Richard Eden, have been predicting tremendous breakouts of GaAs into the digital world. What potentially may be the start of a new generation of merchant vendors is represented in GigaBit Logic, Inc. (a recent sionour of Rockwell's Thousand Oaks Research Center: founded by Fred Blum. This new start-up has the avowed goal of providing high-speed, high-density digital LSI circuits for the computer market. Fred had been joined by Eden and others - most recently Tony Livingston who left Intel to become Vice President of Marketing at Gigabit. Clearly, the company feeld that customer liaison is not necessary and first samples could be on the street in late 1983.
We have consistently maintained that there has to be a performance/cost benefit of at leas a factor of four or five for a new technology to supplant an older, mature one. It's true that under low-field conditions, electron mobilities in GaAs FETs are as much as six times higher than in NMOS silicon devices. At higher fields, GaAs devices have exhibited velocity saturation and effective mobility has dropped to about twice that of silicon.
GaAs may, however, have the potential for extending its performance well beyond the point where silicon "runs out of gas" (pardon the pun). It has been reported that in layers formed by molecular beam epitaaxy, the mean free path of carriers is greater than device length and thus frift velocity is not effected by the crystal lattice or impurities. Still, it appears difficult to realize the full potential gain. High speed LSI digital performance may be quite elusive due to on-chip interconnection delays and off-chip translation drivers.
It has been very difficult to compare performance of the circuit milestones of GaAs to those of silicon. The initial development of GaAs integrated cuituits migrated from the analog applications and raw speed was the goal. Submicron lithography and refractory and noble metallization (with the attendant high investment in e-beam, direct setp on wafer, ion milling, etc.) have made comparisons difficult. Also, the majority of circuit implementations, until recently, have been inherently different for those of silicon.
Early work in Mesa structures, which evolved from the early discrete devices where active elements were shaped by etching the GaAs substrate, have given way to present planar implanted citrates resembling silicon integrated circuit devices.
Evidence of difficulty in assessing performance differences was given in an invited paper delivered by Marty Lepselter a year ago at the GaAs symposium. His paper revealed that Bell Labs had fabricated silicon MOS transistors with 0.25 micron channel length and 28 ps. Switching times - performance supposedly available only through GaAs structures.
In the early stages of GaAs development, it was quickly discovered that material research was the crux of the problem and that quality substrates were the key to success. Giving evidence to this is the fact that Harris Corporation selected Dick Soshea, formerly head of optoelectronic development at Hewlett-Packard Company, to heat up its new microwave semiconductor operation. What to light emitting diodes and GaAs microwave devices have in common? The GaAs substrate.
Have innovations and sub-processes of silicon technology finally evolved to the generic stage where they can be "borrowed" by the GaAs aficionados? Have the paths of both silicon and GaAs procesion equipment finally crossed due to the common requirements of submicron geometries? Has the tremendous leverage of the silicon semiconductor instustry provided the resources in basic material understanding? Only recently has a nominal supply of reasonably-high quality GaAs 3-inch wafers become available. However, there are presently only three vendors at that wafer size, with the rest still limited to supplying 2-inch versions. Today, undoped liquid-encapsulated Czochralsk (LEC) semi-insulating material process has taken over from the chrome-doped horizontal Bridgeman (HB) process for crystal growth. Etch pit count and gettering sites, etc. still trail far behind silicon perfection.
GaAs wafer costs are still high compared to silicon: greater by a factor of ten. Due to this high cost, wafer breakage or whole-wafer survival becomes a significant potion of the finished fabricated GaAs wafer cost. Automated cassed-to-casseet handling, becoming quite commonplace in the silicon shop, helps to alleviate the problem in GaAs as well.
GaAs does not have an intrinsic oxide, however this has been circumvented by the use of silicon nitride of oxide deposited in commercial sputtering, plasma-enhanced chemical capor deposition of LPCVD systems, with results comparable to silicon experience. However, surface states at the interface causes inversion, producing leakage and undesired coupling. Epitazal growth of heterogeneous materials, not within the generic mainstream of silicon development, has played a major role in alleviating these parasites. Ion implantation, now standard for impurity introduction in silicon, is almost universally accepted for active-layer fabrication in GaAs as illustrated in the MESFET processing sequence of Figure 1.
[ICE image 7544]
Due to the major push for raw speed, GaAs had always been characterized by very short channel lengths and very fine feature sizes. The 10x step-and-repeat aligners have benefited in the III-V technologies perhaps more significantly than they have affected the mainstream silicon technology. The push for small feature size by both the VHSIC (GaAs was considered not sufficiently marure for inclusion in this program) thrust in this country and the VLSI activities in Japan have helped to force lithography and other requirement of fine-line geometry to process capabilities in the 1-micron region.
Dry processing, particularly for refractory metals such as titanium and tungsten, is quite compatible with GaAs technology. Plasma etching, reactive-ion etch, reactive-ion beam etching and ion million have the potential to be suitable and compatible with the requirements for fine feature size of both silicon and GaAs. There has been considerable development of enhanced lift-off techniques for fine-line pattern replication. Previously, these have exhibited poor yield and were considered unreliable at very small feature sizes. The majority of these processing developments, first developed for the advanced silicon technologies, have been found to be quite applicable to the demands of GaAs fabrication. GaAs has promoted the Schottky barrier contact and has been able to leverage off similar developments in silicon. Magnetron sputtering systems have provided a versatility of metallication depositions; double-layer metal with polymide interlayer dielectric - perfect for GaAs use, as well.
The rule of 5:1 performance advantage required for a competing technology general applies in computer design where tradeoffs are made between LSI technology and design, packaging methodology, hardware architecture and software. However, a significant performance advantage coung be a few tenths of a dB in noise figure or a factor of two increase in cutoff frequency or bandwidth of operation. In many analog applications, the performance rests solely on the shoulders of the technology chosen.
Early applications of GaAs involved on-for-one substitution for discrete silicon transistors in low-noise microwave amplifiers. GaAs has continued to be very pervasive in these and similar niche implementations; moving from primary use in low-power front-end circuits into modulators, mixers, and multiplexers as well as into higher-power applications. The potential of improved life and reliability over existing beam amplifiers such as traveling wave tubes (TWTs) and klystroms can provide almost a factor of 10 decrease in cost of ownership, when power supplies and other equipment are considered. However, it must be recognized that this is a limited application market.
GaAs devices have performed extremely well in niche applications as discrete devices incorporated into hybrids or even ions SSI form for high-frequency analog microwave circuits.
The most promising result from the research and development laboratories has been the evolution of monolithic microwave integrated circuits (MMICs), arising from the well-understood foundations of hybrid design. See Figure 2. Many people thought we were rapidly moving down the learning curve when the MMIC was developed. However, all this did was to integrate some of the discrete circuit components (including passive devices found in the hybrid) onto one chip, enabling greater interconnection - but not affording any greater level of overall integration than SSI.
The present GaAs device market is extremely fragmented, as a majority of suppliers focus on very specialized microwave applications. Approximately 50 percent is military-oriented, with most of the remainder in telecommunications. No one company appears to have more than 15 percent of the market. Most companies manufacture about 50 percent for the open market (often sold to competitors), with the remainder destined for their own use. Due to the specialized nature of the market, user companies encourage the development of competitive parts in order to ensure a second source.
An application that can have an overwhelming effect on GaAs technology is the use of loise noise amplifiers for home satellite receptions (also termed the direct broadcast satellite, or DBS, market). With 70,000,000 dwelling unites in the U. S. alone, a number of new DBS receiver manufacturers will emerge, but will have high risks due to insufficient material support.
There is considerable dohotomty between the present development activities and the expected volume markets. There is a wide range of specialized chips required for both the high-cost military market (with an annual volume of maybe a million pieces with "affordable" prices of $100 and up) and the potentially larger market for DBS front-end chips (with an annual market of 10 to 100 million chips at prices in the range of $10).
The speed advantage of GaAs is certainly important, but one must also consider the cost advantage of the potentially-lower power dissipation of GaAs circuits. It has been estimated that an Amdahl V470 mainframe, if placed on a single chip of GaAs would have an average gate clocking frequency of 1 gigaHertz, five times the present performance, but would only consume 1/100 the power of the present mainframe. The implications of power supply cost reductions are significant.
A very fast cache memory would be a potential high-volume digital component. A pin-compatible GaAs cache RAM that could fill existing bipolar RAM "sockets" could readily increase performance of large mainframes, with only insignificant design changes required. In order to complete with bipolar devices, however, an access time of approximately 1 nanosecond for a 1K or 4K memory chip is needed. Since bipolar devices of that density are now exhibiting access times in the 5- to 10- second nanosecond range, is it worth the estra effort of levelshifting, etc. to be compatible for the processors ECL outputs and interconnections? Fast GaAs RAMs have not yet been sampled on the market (although 1- to 2-ns accesstime devices have been reported by several companies). The momentum of silicon is continuously eroding the performance gap!
There have been a number of outstanding digital MSI circuits (and even what could be considered LIS circuits) developed, such as Rockwell's 8X8 multiplier. First announced several years ago, the Rockwell device reportedly computes a produce in 5 nanoseconds. Such GaAs components are at least two or three years away from commercial feasibility. However, this and similar research circuits have had good press coverage, but are not indicative of the general state of the digital integration art with respect to GaAS. Large companies with significant market driving forces - such as the high-end microwave instrumentation of H-P and Tektronix or the military aerospace applications of Rockwell and Highes - have afforded the research and development for specialized components in those aresas. IBM, a company that could well afford the investment in GsAs digital technology, has chosen Josephson Junctions as the "horse to ride."
The market in GaAs overall is expected to grow at a rate comparable to that of silicon over the next decade. This is not an explosive growth. Digital applications for GaAs are expected to be equal to analog in market size by 1990.
A potential of application is
in optical interconnection. The very high band-width achievable in light
pipes (fibre optics) are compatible with the light emitting and sensing
properties of III-V materials and would provide an excellent marriage
of technologies at high gate counts and high I/O counts. Figure 3 is
one novel approach using a well approach to laser fabrication.
| The Archives Center maintains container
lists with unique collection names, series numbers and most important
the Archive Center Number. When requesting information for your research
either from the collecting division or the Archives it is imperative to
include the associated number(s). Your contacts for requesting information
and/or appointments to research the Chip Collection are:
Archives Center - Manuscript Technology Collections
Hal Wallace, Curator - CHIPS
Information Technology & Society - Electricity Collections