Issue

Semiconductor manufacturing and vacuum technology: A memoir

05/01/1997

Semiconductor manufacturing and vacuum technology: A memoir

Robert K. Waits, UTI Division of MKS Instruments Inc., Santa Clara, California

Of the earliest semiconductor manufacturing processes, only two required a vacuum. Today, a quarter of high-performance logic process steps are performed under a vacuum. Vacuum chambers have evolved from bell jars serviced by diffusion pumps to ultra-high vacuum chambers using turbomolecular pumps. As critical dimensions continue to decrease, cleaner, more efficient, and lowerpressure vacuum environments are becoming the norm.

Back at the dawn of the IC industry, circa 1960, only two manufacturing process steps required a vacuum: the evaporation of aluminum for the one-and-only metal layer and the evaporation of gold on the back of the silicon wafer so that the circuit die (now called a chip) could be solder mounted in a gold-plated, TO-5 transistor package. The vacuum chamber was a rubber-gasketed glass bell jar pumped by an oil diffusion pump, possibly filled with the latest new Dow-Corning silicone pump fluid, which didn`t turn to brown goop if it was accidentally exposed to air (Fig. 1). The high-vacuum gauge was of the Bayard-Alpert design invented just ten years before. The aluminum was evaporated from a tungsten filament (probably bought from R. D. Mathis). No problem with thickness control, you just counted out the correct number of aluminum "clips," loaded them on the filament, and evaporated to completion. Much of this technology was borrowed from the telescope-mirror makers and the vacuum tube industry.

Figure 1. A typical 18-in. bell jar used for metal evaporation, ca. 1961. Ionization and thermocouple gauges with stainless-steel flanges were connected to the system with metal gaskets. (Photo courtesy of CVC Products)

By 1968, several companies (Thermionics, Temescal, Sloan, Varian) offered a new way to evaporate: the magnetically focused electron-beam gun. At about the same time the quartz crystal oscillator thickness monitor became commercially available (Sloan) and, together with the electron-beam gun, provided a technological breakthrough. The 270? deflection electron-beam gun (Temescal), with the electron- (and contaminant-) emitting filament hidden from the substrate, arrived just in time to provide the sodium-free aluminum that MOS transistors required. A tungsten-filament evaporator couldn`t do it. At that time (1965-70), most evaporation systems were do-it-yourself projects assembled from off-the-shelf components, for example, a diffusion pump from NRC or CVC, valves from Temescal and Veeco, gauges from Veeco, CVC, and NRC, a Temescal electron-beam gun and a 15-kV power supply to run it, a Welch mechanical pump and other essential odds and ends, including a heated rotating planetary substrate holder from Carl Hermann (now CHA Industries).

There were those who felt that diffusion-pumped systems were "dirty" and the more advanced systems had a liquid-nitrogen-cooled trap (CVC, NRC, Davis & Wilder, or Granville-Phillips) with an anti-creep baffle above the diffusion pump to prevent contamination. Later it was found that most of the contamination came from the mechanical pump during rough pumping. This was discovered using one of the small magnetic sector mass spectrometers (residual gas analyzers or RGAs) that had become available in 1960 (from CEC). But the liquid nitrogen (LN2) trap was a great pump for water vapor, which the RGA identified as the major residual gas.

Oil-free high-vacuum systems with metal-sealed Varian (Wheeler) flanges and Varian or Ultek ion pumps were tried. Rough pumping from atmosphere was accomplished with liquid-nitrogen-cooled canisters filled with Linde molecular sieve. However, even if these systems were cleaner (and this was controversial), their higher cost and lower pumping speed prevented their wide acceptance. Anyway, fluorocarbon elastomer seals had become available that mitigated the hydrocarbon contamination problem a bit - and, they could be baked at 125?C. Now (1997) everyone also wants "dry" pumps to rid the process of hydrocarbons.

What about sputter deposition (now called PVD by the marketeers)? It was pioneered as an electronic component production process by Western Electric for depositing tantalum and tantalum nitride resistors on hybrid circuits in the first Touch-Tone phones. Sputtering was used in the early 1970s for depositing platinum silicide contacts and gold beam-leads on ICs at IBM and elsewhere, and for forming nichrome or silicon-chromium thin-film resistors on analog ICs. Radio-frequency (RF) sputter systems became available in the early 1960s (from R. D. Mathis). RF sputtering held the promise, never fulfilled, of being a means to deposit an insulating layer of silicon dioxide on top of aluminum so that a (gasp) two-layer IC could be manufactured. But sputtered silicon dioxide films were prone to cracking and ordinary atmospheric-pressure chemical vapor deposition won out. Later, RF sputtering proved extremely useful for sputter cleaning prior to film deposition and its relative, RF-sputter etching, is used to fabricate the read/write heads that are in all computer disk drives.

The next vacuum application that came along (around 1969) was ion implantation, pioneered commercially by High Voltage Engineering, Accelerators Inc., and Extrion (now part of Varian). At first, no one was sure what kind of vacuum pumps to use or exactly where to put them: end station, beam line, or ion source? Having the ion source at 100 keV did not simplify matters. This situation is still being sorted out.

In 1974, the rediscovered magnetron sputter cathode revolutionized sputter deposition. Key players included Peter Clarke, Sloan, Temescal, MRC, and Varian. For the first time, aluminum metallization for ICs could be sputter-deposited. Conventional glow-discharge sputtering was too slow and the deposited aluminum film was of poor quality. Sputtering now enabled the deposition of the Al-1%Si-2%Cu alloys that were required for electromigration resistance. (Workers at Hughes had patented planar and post-magnetron sputter cathodes in 1965 and had even presented a paper at the 1963 AVS Symposium in Boston, but no one took notice.) 1974 was also the year of the first commercial He-refrigerated cryogenic vacuum pump (CTI) (Fig. 2). This device has supplanted the diffusion pump as the pump of choice for high-vacuum applications such as ion implant, evaporation, and sputter-deposition systems.

Figure 2. Circa 1974, the CTI-Cryogenics Cryo-Torr High Vacuum Pump was one of the first commercially available croypumps used for high-volume semiconductor production. The pump was demonstrated on the Alpha 64K chip production line at IBM in the early 1970s and went on to become the semiconductor industry standard for high-vacuum pumping. (Photo courtesy of CTI-Cryogenics, a division of Helix Technology Corp.)

In the mid-1970s, three new processes required a vacuum atmosphere: low-pressure chemical vapor deposition (LPCVD) (various manufacturers), plasma-enhanced chemical vapor deposition (PECVD) (Applied Materials), and plasma etching and resist stripping (Tegal, LFE). LPCVD was needed to increase manufacturing throughput. A pretty simple operation: you just hung a vacuum pump on the back end of a diffusion tube and ran reactive gases into the front end. However, the pressure was in the torr range rather than the microtorr and millitorr ranges used in evaporation and sputter deposition. At first, mechanical pumps were used, but new (at least to semiconductor manufacturing) pumping systems that could gulp larger quantities of gas were needed: a Roots blower backed by a mechanical pump. But the pumps were corroded and the pump oils were turned to acid-laden sludge by the exhaust products and particles from LPCVD reactions. Filters, acid-neutralizers, and finally, very expensive fluorocarbon pump fluids were required to solve these problems. About the same time, plasma etching
esist-stripping and plasma-enhanced CVD were gaining notice, requiring the same acid-resistant pumps as LPCVD. Newer, lower-pressure, CVD and plasma-etch applications require modern turbopumps. A better pressure-measuring device was necessary for the torr pressure regime and the capacitance manometer was just the thing (MKS).

Vacua are hidden in some other semiconductor manufacturing processes. Masks are created by electron-beam writing, which, of course, requires a vacuum. Priming of wafers with hexamethyldisilazane prior to photoresist spin is often done in a vacuum.

As the semiconductor industry grew, the manufacturers of vacuum components started to offer complete manually operated, and then automated, vacuum evaporators and sputter systems. Other companies started from scratch and designed and built systems for ion implantation, plasma deposition, and etching; they furnished the system, but the customer had to develop the process. Today, of course, the semiconductor industry demands a turn-key system with a process guaranteed to produce 100 (or more) perfect wafers/hr (see "Milestones in semiconductor-related vacuum technology" on p. 111).

Whither vacuum?

Vacuum environments of the 21st-century will have to be lower pressured, cleaner, and achieved faster than those of today. To no-one`s surprise, cleaner vacua - with no reactive gases such as water and oxygen - promote the formation of denser films and allow high-quality, epitaxial films to grow at lower temperatures. Low-pressure (<5 mtorr), high-density plasma etch, and PECVD systems require high pumping speeds to get low-residence times for etchants and precursors. More efficient and reliable turbomolecular pumps are coming to the rescue. If x-ray, extreme UV (13-?m) or possibly deep UV (193-?m) lithography becomes a manufacturing reality, a vacuum or helium atmosphere will be required during resist exposure. Other vacuum processes, such as molecular beam epitaxy (MBE), may make it out of the lab and into production.

At first, only two process steps involved a vacuum environment. Today, of the 363 steps listed in the SEMATACH high-performance logic process sequence, approximately 25% require a vacuum. Perhaps a corollary of Moore`s law is at work that predicts that by the time we get to molecule-wide critical dimensions, all processes will be done in a vacuum.

Acknowledgment

This, being a "memoir," is subject to the failings of remembrances of things past, and the author apologizes for any errors and omissions concerning dates, players, and those first to market.

Reference

1. Dictionary of Vacuum Science and Technology, ed. M. Kaminsky and J.M. Lafferty, American Vacuum Society, New York, 1980.