How IBM invented semiconductor manufacturing automation

In 1970, Bill Harding envisioned a totally automated wafer-fabrication line that might produce built-in circuits in lower than someday. Not solely was such a objective gutsy 54 years in the past, it could be daring even in at the moment’s billion-dollar fabs, the place the fabrication time of a complicated IC is measured in weeks, not days. Back then, ICs, similar to random-access reminiscence chips, have been usually produced in a monthlong stop-and-go march by dozens of guide work stations.

At the time, Harding was the supervisor of IBM’s Manufacturing Research group, in
East Fishkill, N.Y. The mission he would result in make his imaginative and prescient a actuality, all however unknown at the moment, was referred to as Project SWIFT. To obtain such an amazingly quick turnaround time required a stage of automation that might solely be completed by a paradigm shift within the design of integrated-circuit manufacturing strains. Harding and his crew completed it, reaching advances that might finally be mirrored all through the worldwide semiconductor trade. Many of SWIFT’s groundbreaking improvements at the moment are commonplace in at the moment’s extremely automated chip fabrication vegetation, however SWIFT’s extremely quick turnaround time has by no means been equaled.

SWIFT averaged 5 hours to finish every layer of its fabrication course of, whereas the quickest fashionable fabs take 19 hours per processing layer, and the trade common is 36 hours. Although at the moment’s built-in circuits are constructed with many extra layers, on bigger wafers the scale of small pizzas, and the processing is extra advanced, these components don’t altogether shut the hole. Harding’s automated manufacturing line was actually, really, swift.

A Semiconductor Manufacturing Manifesto

I encountered Harding for the primary time in 1962, and hoped it could be the final.
IBM was gearing as much as produce its first fully solid-state laptop, the System/360. It was a considerably rocky encounter. “What the hell good is that?” he bellowed at me as I demonstrated how tiny, unpackaged semiconductor cube may very well be robotically dealt with in bulk for testing and sorting.

Author Jesse Aronstein [at far right, in top photo] took a break from managing the gear group of Project SWIFT to play French horn one night per week with the Southern Dutchess Pops Orchestra. Another key supervisor, Walter J. “Wally” Kleinfelder [bottom left], standing at proper, headed the method group of Project SWIFT. William E. “Bill” Harding [bottom right], seen right here in 1973, was a brusque WW II fight veteran and artistic innovator. He conceived and directed IBM’s Project SWIFT, which succeeded in fabricating built-in circuits in someday.Clockwise from prime: IBM/Computer History Museum; IBM (2)

William E. (“Bill”) Harding was an modern thinker and inventor. He had been creating semiconductors and their manufacturing expertise at IBM for 3 years when the corporate’s new Components Division was shaped in 1961. Harding grew to become a midlevel supervisor within the new division, accountable for creating and producing the gear required to fabricate the System/360’s solid-state units and circuit modules.

He was tough across the edges for an IBM supervisor. But maybe it was to be anticipated of somebody who had grown up in Brooklyn, N.Y., and was wounded thrice in fight in World War II whereas serving in General George S. Patton’s Third Army. After the conflict, Harding earned bachelor’s and grasp’s levels in arithmetic and physics and have become a member of IEEE.

I joined IBM in 1961, coming from rocket-engine improvement at General Electric. Like most engineers on the time, I knew nothing about semiconductor manufacturing. Five years prior, I had attended a vacuum-tube electronics course by which the professor described the transistor as “a laboratory curiosity, which may or may not ever amount to anything.”

Project SWIFT occupied a small house, proven right here in yellow, in constructing 310 at IBM’s sprawling East Fishkill semiconductor facility. IBM

Harding’s tough and crude method surfaced each time I crossed paths with him. If he ever went to IBM
“charm school” (administration coaching), there was no discernible proof of it. Nevertheless, he succeeded in his mission. By 1964, solid-state logic modules for System/360s have been flowing from the Components Division’s new facility on a former farm in East Fishkill.

In July 1970, I returned to IBM after three years of graduate examine. I used to be a first-level supervisor for 4 years previous to that academic break, and didn’t need one other administration job. I wished a purely technical profession, and I joined East Fishkill’s Manufacturing Research (MR) group hoping to get one.

Harding and I then crossed paths once more. In mid-August of 1970, he grew to become MR’s prime supervisor. Prior to that, he spent a yr creating an IBM company technique for the long run manufacturing and use of
very-large-scale built-in (VLSI) circuits. He was given command of MR to reveal the viability of his manufacturing ideas.

An meeting of MR personnel was convened to announce the administration change. After being launched, Harding described his view of future VLSI purposes and manufacturing. These have been his key factors:

• VLSI circuits could be based mostly on field-effect transistor expertise (on the time, bipolar-junction transistors have been dominant);
• Defect-free excessive yields could be paramount;
• Manufacturing could be absolutely automated;
• Best outcomes would accrue from processing one wafer at a time;
• Short turnaround occasions would confer vital advantages;
• Volume would scale up by replicating profitable manufacturing strains.

After the tutorial lecture, Harding modified from professor to commander, General Patton–fashion. MR’s sole mission was to reveal Harding’s concepts, and ongoing initiatives not aligned with that objective could be transferred elsewhere inside IBM or deserted. MR would show that an automatic system may very well be constructed to course of about 100 wafers a day, separately, with excessive yield and a one-day turnaround time.

What? Did I hear that proper? One-day turnaround from naked wafer to completed circuits was what we might now name a moon shot. Remember, on the time, it usually took greater than a month. Did he actually imply it?

Harding knew that it was theoretically potential, and he was decided to realize it. He declared that IBM would have a considerable aggressive benefit if prototype experimental IC designs may very well be produced in a day, as an alternative of months. He wished the circuit designer to have testable circuits the day after submitting the digital description to the manufacturing line.

One-day turnaround from naked wafer to completed circuits was what we might now name a moon shot.

Harding instantly organized an gear group and a course of group inside MR, naming me to handle the gear group. I didn’t wish to be a supervisor once more. Now, reluctantly, I used to be a second-level supervisor, accountable for creating all of the processing and wafer-handling gear for a yet-to-be-defined manufacturing line that I had barely began to visualise. My dream analysis job had lasted little greater than a month.

Walter J. (“Wally”) Kleinfelder transferred into MR to handle the method group. They would choose the product to fabricate and outline the method by which it could be made—the detailed sequence of chemical, thermal, and lithographic steps required to take a clean silicon wafer and construct built-in circuits on its floor at excessive yield.

Kleinfelder chosen a random-access reminiscence chip, the IBM RAM II, for our demonstration. This product was being produced on-site at East Fishkill, so we might have all the pieces we would have liked to construct it and consider our outcomes relative to these of the present nonautomated manufacturing line.

IBM’s SWIFT Pilot Wafer Fab Had a Monorail “Taxi”

Integrated-circuit manufacturing entails first creating the transistors and different parts of their correct locations on the silicon wafer floor, after which wiring them collectively by including a skinny movie of aluminum selectively etched to create the required wiring sample. That skinny movie of conductor is called the wiring, or metallization, layer.

IC manufacturing makes use of
photolithography to create the numerous layers, every with a particular sample, wanted to manufacture an IC. These embrace the steel wiring layers, of which there could be greater than a dozen for a complicated chip at the moment. For these steps, the steel layer on the wafer is coated with a light-sensitive photoresist materials, after which a picture of the sample is uncovered on to it. The areas the place conductors can be shaped are blocked from the sunshine. When the picture is developed, the resist is faraway from the sample areas that have been uncovered, enabling these areas to be etched by an acid. The remainder of the floor stays protected by the acid-proof resist. After etching is accomplished, the remaining protecting resist is eliminated, leaving simply the wiring layer within the required sample.

The IC course of additionally makes use of lithography to create transistors and different parts on the silicon wafer. Here, openings are etched in insulating layers by which tiny quantities of particular impurities could be infused into the uncovered spots of pure silicon to alter {the electrical} properties. Producing the RAM-II ICs required 4 separate lithographic operations utilizing 4 completely different patterns: three for creating the transistors and different parts, and one to create the steel wiring layer. The 4 patterns needed to be precisely aligned with each other to efficiently create the chips.

Lithography is just a part of the IC manufacturing course of, nevertheless. In the present manufacturing line, it took many weeks to course of a RAM-II wafer. But the uncooked course of time—the time a wafer spent really being labored on at numerous thermal, lithographic, chemical, and deposition stations—was lower than 48 hours. Most of a wafer’s time was spent ready to bear the subsequent course of step. And some steps, chemical cleansing particularly, may very well be eradicated if wafers progressed shortly from one step to the subsequent.

It was the duty of Kleinfelder’s group to find out which steps may very well be eradicated and which may very well be accelerated. The ensuing uncooked course of time was lower than 15 hours. It then fell to
Maung Htoo, my supervisor of chemical-equipment improvement, to check the proposed course of. His folks hustled 1.25-inch-diameter wafers by a “pots and pans” lab setup to guage and refine it. The abbreviated process efficiently produced working circuits in about 15 hours, as anticipated.

The structure of an automatic system materialized. It was initially envisioned as a collection of linked machines, every performing one step of the method, like an car meeting line. But gear downtime for preventative upkeep and restore of breakdowns needed to be accommodated. This was achieved by the insertion of short-term storage “buffers” that might quickly retailer wafers at chosen factors within the course of chain when mandatory.

This course of chain idea was additional disrupted by issues associated to lithographic-pattern imaging. Exposure of the photoresist on wafers was generally completed on the time by a course of analogous to photographic contact printing. The lithographic masks, by which gentle shone when exposing the photoresist, was the equal of a photographic adverse. Any defect or particle on the masks would lead to a corresponding defect on a chip, on the identical location, wafer after wafer.

The East Fishkill lithography group had developed a noncontact 10:1 discount
step-and-repeat picture projector. Think of it as a kind of photographic slide projector that produced a shrunken picture containing the sample for a single layer on a chip. It then “stepped” throughout the wafer, exposing one chip location at a time. Relative to contact masking, the stepper promised decrease sensitivity to particulate contamination, as a result of the scale of the shadow of any stray particle could be decreased by 10:1. Other benefits included greater optical decision and longer masks life.

– YouTubewww.youtube.com

Because it was gradual, although, a number of steppers could be wanted to satisfy the throughput goal. Achieving the perfect sample alignment on every wafer for a number of sample exposures required {that a} wafer be routed again to the identical stepper for publicity of every layer within the course of chain. That would cancel the impact of picture distortions launched by slight variations from one machine to a different. Building the RAM-II circuits then required {that a} wafer make 4 separate journeys to its assigned stepper. That divided the linear sequence into 5 sectors. A monorail “taxi” would take a wafer from one processing sector to its assigned stepper, and return later to take it to its subsequent sector.

Each of the 5 sectors was envisioned to be an enclosure containing all the automated wafer-processing and dealing with gear required to perform that section of the method chain. The sector enclosures and the taxi could be designed to offer a clean-room-quality native setting for the wafers. Within a sector enclosure, usually, a wafer would move straight from a wet-chemistry module to miniature furnaces to a photoresist software module, and, lastly, to the taxi pickup port. Inside the wet-chemistry module, for instance, the wafer would bear cleansing, improvement of the photoresist and its elimination, and etching, amongst different procedures.

Control of the whole line was to be completed at three ranges. Overall production-line administration, recordkeeping, taxi logistics, and course of monitoring could be dealt with by a central computer-based system. Dedicated controllers, one for every sector, would handle wafer logistics inside the sector and feed wafer site visitors and processing knowledge to the central system. The particular person processing and wafer-handling modules inside every sector enclosure would have their very own specialised controls, as wanted, for impartial setup and upkeep.

Finally configured, our automated demonstration line for the RAM-II chips would consist of 5 sectors, a taxi, and a lithographic-pattern imaging middle, all managed by laptop. Six months after Harding took command, MR began to design and construct the precise system.

The Brash Middle Manager Found Inspiration in Literature

Harding made frequent journeys to
IBM’s headquarters, in Armonk, N.Y., to report progress, request assets, rebut challenges, and persuade the highest brass that the cash being spent was a superb funding sooner or later. It was a tricky mission. His prolonged weekly employees conferences typically mirrored the strain he was beneath. He lectured at size on issues he knew we knew, informed allegorical tales, and spun analogies.

At the time, I didn’t notice that he was utilizing his employees conferences to develop and refine concepts for the shows at Armonk. He was noting our reactions and adjusting his presentation concepts accordingly. His shows to the highest brass have been efficient. For the period of the mission, spanning about three years, MR had all of the funding and help it wanted to develop, design, construct, and function the whole system.

At one employees assembly, Harding learn aloud Heywood Broun’s quick story “
The 51st Dragon,” to emphasise the facility of a reputation or slogan to inspire folks to realize the unattainable. His level, after all, was that we would have liked a extremely good identify for the mission. “SWIFT” was finally chosen. Harding all the time insisted that it was not an acronym, however however folks figured it was shorthand for “Semiconductor Wafer Integrated Factory Technology.”

SWIFT’s extremely quick turnaround time has by no means been equaled.

SWIFT’s processing and wafer-handling gear was customized completely inside IBM’s Components Division. The major design goals have been to course of wafers robotically, constantly, and uniformly and hold them clear and undamaged. Wafer-handling experiments sorted out the cleanest and gentlest methods. Handling gear was designed to help the wafer reasonably than grip it. A novel wafer handler that used a circulate of air above the wafer to raise it, with out bodily contact, was efficiently included for a few of the wafer-transport strikes.

There was one exception to the “clean and gentle” design of SWIFT’s dealing with equipment. Management on the Components Division’s Burlington, Vt., website pressured Harding to make use of “air-track” wafer-transport gear that that they had developed. This gear used airflow to raise and transfer wafers, very similar to a puck in a sport of air hockey. Harding wanted Burlington’s continued help, so he decreed that some air-track gear be utilized in SWIFT. And it was, although wafer-contamination and reliability questions have been unresolved.

Another top-down decree explains why SWIFT ended up with two several types of sector management programs—the antithesis of fine design for maintainability. A customized controller had been designed, and 5 models have been being constructed (one for every sector), when HQ required that we incorporate the newly introduced
IBM System/7, which had been developed particularly for factory-equipment and process-control purposes. After all, if IBM itself didn’t use the pc in its personal superior manufacturing line, potential prospects would surprise “why not?” But if SWIFT used a System/7 and the mission proved to achieve success, it could assist promote System/7s. And so for the 5 sectors, SWIFT ended up with 4 customized controllers and one System/7. Both varieties labored properly.

Equipment reliability was SWIFT’s Achilles’ heel. To assist obtain excessive reliability and ease of upkeep, sure mechanisms and controls have been standardized to be used all through the system, they usually have been chosen for reliability and ease reasonably than novelty or magnificence. For instance, an individual observing the system in operation would discover that many motions have been completed in discrete clean steps reasonably than a single traverse. Underlying that peculiarity was the in depth use of the easy, sturdy, and dependable
Geneva drive, initially developed centuries in the past for clocks, however now tailored for linear and rotary motions that needed to be clean and exactly locked in on the finish factors. Each simply managed flip of the Geneva drive’s enter shaft made one step. Long traverses required a number of turns of the shaft, ensuing within the odd-looking motions.

Inside a sector’s enclosed chamber, a wafer went by a collection of completely automated processing steps. Two of the early idea sketches are represented right here. The wafers got here into the higher chamber with a sample uncovered onto the resist and underwent a collection of processing steps that included improvement, hardening, etching, and others, as indicated.

Another simplification concerned spinning the wafers to centrifugally unfold liquid photoresist that was dropped onto the middle of the wafer. In present strains, “wrong spin speed” was regularly cited as the reason for resist-related wafer-processing rejects. Spin pace was eradicated as a variable by driving SWIFT’s spinners with synchronous AC motors locked to three,600 rpm by their 60-hertz AC energy supply, simply as phonograph turntables are pushed. No pace controllers could be required. The desired photoresist movie thickness could be achieved by adjusting the remaining variables—temperature, viscosity, and/or spin time. In the top, system reliability was improved by the elimination of 4 separate pace controllers.

As SWIFT progressed from blue-sky idea to precise {hardware} implementation, Harding adjusted MR’s group and gained the cooperation of supporting teams. He noticed to it that his folks had the assets to do the job and will deal with the mission. I got here to admire his organizational expertise and his capacity to single out and recruit top-notch expertise from inside the firm.

Harding established a bunch to develop SWIFT’s grasp management system, which monitored the progress on each wafer because it moved by the sectors. This Execution Control System (ECS) was based mostly on an
IBM 1800. Each wafer had a serial quantity and was tracked at each step by the road. The ECS saved and monitored every wafer’s processing parameters, detecting and reacting shortly to out-of-spec conditions. Its punch playing cards and tape cartridges appear quaint by at the moment’s measure, nevertheless it was a serious advance in manufacturing management and monitoring for a wafer line.

He additionally transferred a complete instrumentation division, managed by Sam Campbell, from IBM Endicott to East Fishkill. Campbell’s division subsequently developed groundbreaking strategies for real-time, in-situ course of management for SWIFT.

A Short Life however an Enduring Legacy in Semiconductor Manufacturing

Mockups of furnaces and chemical processors have been constructed and examined.
Robert J. Straub’s division in East Fishkill’s Manufacturing Engineering group designed and constructed the sectors and the processing gear modules inside them. Harding introduced in Bevan P.F. Wu to handle the set up, debugging, and operation of the road. As gear and services coalesced in SWIFT’s devoted 4,000-square-foot house, Rolf H. Brunner, who had managed a superb portion of the sector designs together with improvement of the vacuum metal-deposition gear, took duty beneath Wu for gear set up, startup, and debugging.

Only one operation in the whole course of was not absolutely automated. Alignment of the wafer for exposing the sample on the photoresist nonetheless relied on a well-trained operator. In its closing type, SWIFT had each a ten:1 optical stepper and likewise a 1:1 contact-mask machine, however because it occurred, a lot of the chips produced have been with the 1:1 machine, as a result of the throughput was greater that method.

By the top of 1973, IBM HQ was already satisfied that full automation of wafer processing might succeed. So a lot in order that this objective was adopted as a major goal for a brand new wafer-processing line to provide the circuits for IBM’s next-generation laptop, the “FS” (
Future System). The proposed new line was dubbed “FMS” (Future Manufacturing System), and SWIFT was renamed “FMS Feasibility Line.”

Bevan Wu efficiently managed the road’s completion, take a look at runs, personnel coaching, and refinements of kit, course of, and procedures. He introduced the road to the purpose of being certified to provide circuits for IBM merchandise. The system made 5 continuous-operation runs between mid-1974 and early 1975. Between runs, his group analyzed outcomes and carried out enhancements. The longest steady run spanned 12 days. Wafer throughput averaged 58 wafers per day, 83 % of its designed most. Average turnaround time from bare-wafer enter to testable-circuits output was about 20 hours. The uncooked course of time was 14 hours. The yield in the end equaled the perfect ever achieved by East Fishkill’s typical RAM-II manufacturing line.

A complete of 135 technicians, engineers, and managers from IBM places worldwide have been skilled on the operation of the system. They produced 600 product-quality wafers with 17,000 RAM-II FET reminiscence chips.

But like his WWII commander, General Patton, Harding was bypassed to guide “the big show”—in Harding’s case, the creation of the brand new FMS automated line. Leaving the administration profession ladder behind, he was promoted to IBM Fellow, the very best nonmanagement stage within the firm.

The FMS Feasibility Line, initially SWIFT, made its final steady run in early 1975. It had completed its goals. Its folks have been now wanted to assist create the FMS line to provide FS computer systems. But later in 1975, the FS mission was canceled, and FMS grew to become superfluous. A portion of the gear destined for FMS grew to become East Fishkill’s
QTAT (Quick Turn Around Time) line, a groundbreaking IBM showpiece that’s higher remembered than its obscure predecessor, Project SWIFT.

Although SWIFT’s life was quick, and it was by no means within the limelight, its many inventions are clearly seen in at the moment’s semiconductor fabs. Like SWIFT, these fabs are extremely automated and laptop managed; have a central transport system and “Bernoulli” handlers, which exploit the circulate of air to raise wafers with out making bodily contact; apply resist instantly after oxide or steel movie formation; use steppers for lithographic sample publicity; and make use of real-time course of management. All of those have been groundbreaking options of Project SWIFT 50 years in the past.

The expertise of working beneath Harding on SWIFT for 3 years was, for me, transformative. What had began with trepidation ended with admiration. I’ve come to contemplate Bill Harding a real genius, in his personal method. Spurred on and supported by his distinctive administration fashion, a small group of devoted folks achieved way over anybody initially envisioned. More than even we ourselves thought potential.

We consider the primary achievers in an trade because the “fathers” of the trendy embodiment of their innovations. Edison, Bell, Ford, and the Wright brothers, are generally spoken of this fashion. In that sense, William E. Harding is clearly the daddy of the trendy, automated, billion-dollar fab.