Issue



Silicon Germanium: SiGe for mainstream semiconductor manufacturing


01/01/2000







David C. Ahlgren, Basanth Jagannathan, SiGe Technology Development, IBM Microelectronics, Hopewell Junction, New York

Special Section - Technology Outlook

Silicon germanium (SiGe) is the latest semiconductor-manufacturing buzzword as new applications and announcements appear in wireless and high-performance wired markets. For example, in the second half of 1999, Alcatel announced a 10Gbit/sec SONET transceiver - a high-speed optical data transmission link - built solely on IBM's SiGe IC technology. While manufacturers with SiGe capability are still a small part of semiconductor manufacturing, increasingly, participants in the analog segment are developing proprietary process recipes to compete for market share in this lucrative business.

SiGe origins

Over the past 5-10 years, wireless and high-performance wired applications have turned increasingly to gallium arsenide (GaAs) and other III-V compound semiconductor technologies. In 1992, bolstered by more than a decade of research at the Watson Research Center, several individuals within IBM were prepared to develop commercially acceptable analog SiGe technology. The premise for this technical leap was to transform standard silicon (Si) into a heterojunction semiconductor material by doping the Si lattice with germanium (Ge). While there were many process integration issues to address, if successful, this move would make it possible to compete with III-V semiconductor performance, simultaneously maintaining the low cost advantages traditionally associated with Si (i.e., mature production facilities, 200mm substrates, low defect densities, etc.).

The Si-SiGe HBT

Metamorphosis from a Si homojunction to an SiGe heterojunction bipolar transistor (HBT) is made by growing an epitaxial layer of SiGe alloy on an Si wafer after full device isolation has been completed. Basic physics tells us that the atomic size difference between these two species necessarily results in a lattice mismatch that must be dealt with in integrating this layer. While a certain level of strain is required for band gap reduction - a driving force behind HBT performance improvement - the strain must be kept low enough to avoid a generation of lattice defects and unstable SiGe films.

The second, and equally important consequence is the SiGe epitaxy must allow growth of a boron-doped base region of an npn transistor to replace the ion-implanted base conventionally used in bipolar junction transistors. This results in narrow base widths and frees device designers from the confines of Gaussian profiles and implant-channeling effects. The net effect is a dramatic performance boost to the Si device [1].

The SiGe fab

Click here to enlarge image

RIGHT. CMOS process flow with required additions to produce SiGe-based BiCMOS and additional analog devices.

The earliest SiGe commercial process, qualified for production in 1996 at IBM's Advanced Semiconductor Technology Center, revolved around an HBT with a maximum available power gain frequency of 65GHz [2] and various active and passive device support elements, providing designers with a wide palate for high-performance analog ICs. With control of cost as a primary driving force, process integration issues were addressed with the challenge to keep the SiGe process compatible with standard CMOS processing. Early in process development, we recognized that this compatibility would enable use of the SiGe HBT for BiCMOS using a preexisting ASIC CMOS process to merge large CMOS digital logic functions with a high-performance HBT, thus paving the way for high levels of integration not possible with GaAs technologies.

The process flow used for SiGe BiCMOS technology requires a few deviations from normal CMOS processing, principally to add the SiGe epitaxy base HBT (see figure). The additional process cost for transformation from CMOS to SiGe BiCMOS is only ~20%, part of this from inclusion of high-quality passive elements for design flexibility. Within the SiGe BiCMOS process flow, several recipes are "SiGe unique." SiGe epitaxy, however, is the only non-CMOS process equipment requirement.

SiGe epitaxial growth

Because it is desirable to grow the active base region of the npn device within and confine the base to the Ge-doped region to enhance device performance, epitaxial growth temperature must be low (~500°C). Our low-temperature epitaxy (LTE) technique for growing Si and SiGe films originated in the early 1980s with Bernard Meyerson's work on an early ultrahigh-vacuum chemical vapor deposition (UHV-CVD) reactor. Today, 200mm versions of this system, based on IBM designs, are available from Balzers Process Systems and CVD Equipment Corp.

Briefly described, this process starts with a batch of hydrogen-terminated Si wafers, prepared via an HF treatment, in an ultrahigh-vacuum chamber (<=10-11 mbar). Once the wafers are in the UHV chamber, the growth process is started with the introduction of silane. By accurately controlling dopant gas flows, Ge profiles, and associated boron doping, the HBT base is achieved [1].

Within IBM, SiGe UHV-CVD LTE tools are qualified for production at both the Burlington and Hopewell Junction production facilities. One common thread in all the LTE tools based on the IBM design is the emphasis on good vacuum practice and contamination prevention methods. This is what makes the tool robust and productive in a manufacturing environment, despite demanding UHV conditions.

Parallel to IBM's success with SiGe production, other SiGe epitaxial deposition methods were proposed for commercial applications, including molecular beam epitaxy (MBE), low-pressure chemical vapor deposition (LPCVD), and rapid thermal chemical vapor deposition (RTCVD). While some of these systems are commercially available, little data has been published demonstrating tool performance or commercial device characteristics.

Results and future

Resulting from the extremely tight control of the narrow base region, the epi-base transistor has demonstrated unprecedented distributions of DC and AC performance measurements, for example, ±4% control of HBT cutoff frequency across a very large database of 200mm wafers [2], a benchmark unequaled in III-V semiconductor production. With four years of SiGe production experience, chip yields, performance, process control, and reliability have been demonstrated and improved. For example, transistor yields on arrays of 4000 HBT devices average >90% [3], and reliability results show minimal HBT device shifts out to 100,000 power-on-hours [4] with CMOS reliability and yield equivalent to the base ASIC CMOS process [5].

As demands for smaller, lighter, lower-power applications drive the wireless marketplace, a full BiCMOS technology becomes a requirement. This technology allows increasingly higher levels of integration and the combination of ASIC CMOS digital designs with bipolar HBT speed and performance for state-of-the-art analog functions.

IBM is now in its third generation of SiGe technology, which includes 0.40µm and 0.25µm high-performance BiCMOS [6, 7]. Progress has been extremely fast in the introduction of new, world-class semiconductor technology across a broad range of commercial and consumer products. These include cellular basestation and handset components, wireless networking, digital network switching, optical transmit and receiver systems, and global positioning systems (GPSs), such as IBM's recently announced GPS receiver featuring a SiGe front end with direct conversion RF architecture.

In another example, Harris Semiconductor has converted its computer cards for very high bandwidth 802.11 wireless local area networks from Si BiCMOS to IBM's SiGe technology. This resulted in halving the system's chip costs, chip size, and power consumption, and increasing transmission data rate by 550%, and system range 400%.

When asked recently about the future of SiGe technology [8], Meyerson said, "All these products are breakthroughs. In the next 3-5 years, the entire world will move from global systems for mobile communications to code division multiple access with added focus on high-speed data links." To back up his comments, he referred to an SiGe report from Strategies Unlimited that forecasts $1.8 billion of revenue in the year 2005 just from SiGe technology chips. "When you figure what the product worth is, you are probably talking somewhere in the neighborhood of $5-10 billion. This is a breakthrough to me," concludes Meyerson.

References

  1. D.L. Harame et al., IEEE Trans. El. Dev., Vol. 42, No. 3, 1995.
  2. D.C. Ahlgren et al., IEDM Tech. Dig., pp. 859-862, 1996.
  3. S. Subbanna et al., IEDM Tech. Dig., 1999.
  4. D.C. Ahlgren et al., Proc. 28th ESSDERC, pp. 452-455, 1998.
  5. R. Johnson et al., BCTM Proc., 1998.
  6. D. C. Ahlgren et al., BCTM Proc., pp. 195-197, 1997.
  7. S. St.-Onge et al., BCTM Proc., 1999.
  8. M. Helms, Chip: Semiconductor Industry Source Book, 1999, pp. 26-27, (www.bps-IT.com).

David C. Ahlgren received his PhD in chemical physics from the University of Michigan. He is a senior engineer at IBM Microelectronics, Hopewell Junction, NY 12533; ph 914/896-8460, fax 914/892-4604, e-mail [email protected].

Basanth Jagannathan received his PhD in electrical engineering from SUNY, Buffalo. He is a development staff engineer at IBM Microelectronics.