- Buyers Guide
The Road from RFIC to SoC
Twenty years ago, the microwave semiconductor industry in the US and subsequently all over the world got a boost from the Defense Advanced Research Procurement Activity, DARPA’s $600 M MIMIC (Microwave and Millimeter-wave Monolithic Integrated Circuits) program. Driven by an emphasis on production discipline, established through volume production, the industry transformed itself from a supplier of specialty diodes and transistors into a reliable supplier of microwave monolithic integrated circuits (MMIC). Gallium arsenide (GaAs) was the microwave material of choice, clearly superior to silicon for microwave applications due to its higher electron mobility and semi-insulating nature. Significant reviews of the development of this technology have been published, providing comprehensive overviews of the technology development from the days of molecular electronics to the creation of fully monolithic devices.1,2
Although many of the expectations of technical dominance and business riches for GaAs-based semiconductors have not materialized, this technology has enabled the introduction of many breakthrough products that in the last 20 years have changed the way that we live. GaAs transmit/receive (T/R) modules for long-range active array radars warn our troops of incoming threats from missiles or projectiles. GaAs low noise amplifiers (LNA) for direct broadcast satellite TV receivers or cable modems allow us to watch our favorite games wherever they take place. GaAs power amplifiers (PA) and switches for cellular phones connect us wirelessly to our families and our work. Our computers are constantly connected to the world. Whether in our offices or in the coffee shop, we have constant network access. RF semiconductors have enabled the development of disruptive technologies that impact all facets of our lives. This article will focus on some of the major milestones that drove the technical capability and cost of RF semiconductors, enabling this new world to become a reality.
At the beginning of the 1980s, the potential of monolithic integration of microwave functions on semi-insulating GaAs had been well recognized. Impressive demonstrations of the capability of MMIC technology were widely reported.3 Although semiconductor fabrication discipline was established in silicon technology, this discipline was slow to adopt in GaAs processing. The strong commercial pull for silicon-based integrated circuits drove wafer volume, which in turn drove production discipline and learning. Wafer diameters, in silicon production, progressed from three inch to 100 to 150 mm, driven by ever-increasing fabrication (FAB) volume.
In GaAs-based technologies, the material was two inches in diameter. A debate was raging on the advantages of ion implantation versus epitaxy for active layer formation. Processing was dominated by hand dipping in magic solutions in wet chemical benches. Process control monitors (PCM), a key element of FAB control in silicon, were rarely applied. Test was challenging; instrumentation was primitive, methods manual and calibration procedures had not been standardized. Gate lengths were much shorter than in silicon, resulting in significant lithography challenges, but the transistor density for RF applications was very low, yet the MMICs were very large.
Manufacturing discipline in GaAs IC fabrication received its first major push not from RF but from the digital world. In 1982 DARPA initiated the Advanced On-Board Signal Processing (AOSP)4 program focused on high speed digital processing using GaAs. This program was based on the assumptions that state-of-the-art GaAs semi-insulating substrate material was adequate for some types of digital very large scale integration (VLSI) circuits; the low yield of GaAs LSI circuits was primarily driven by random processing defects, not lack of reproducibility in device parameters; strict process control, as practiced in silicon-based VLSI manufacturing, would result in achieving comparable ‘learning curve’ type yield improvements; and a minimum of 100 wafers/week throughput is necessary to achieve pilot manufacturing discipline.4 Pilot lines were established in several companies, many of which were applying GaAs technology to both digital and RF applications. The program, for the first time, began to drive significant wafer volumes through GaAs processing lines.
Following closely on the heels of DARPA’s digital GaAs effort, DARPA launched the MIMIC program with this objective: “Provide the needed microwave and millimeter-wave products at a price that will allow their use in fielded Department of Defense systems, that meet all required electrical, mechanical, and environmental parameters, and that continue to operate reliably for the time necessary to fulfill their intended application.”5
The MIMIC program, building upon the issues identified in AOSP, recognized that there were many areas that required attention to achieve the program goals. All aspects of MMIC fabrication, from materials, to device processing, to test, to design needed to be addressed. The MIMIC program was structured as a multi-phase effort: Phase 0 – definition phase (1987); Phase I – first hardware development phase (1988); Phase II – second hardware development phase (1991); and Phase III – focusing on critical technology development. This comprehensive program, in an effort to drive down the cost of MMIC production and facilitate their use in systems, addressed: (1) the high cost of the starting materials; (2) the poor production control of active layer formation by ion implantation or epitaxial growth; (3) the lack of a comprehensive computer-aided-design systems with appropriate circuit models; (4) the lack of adequate production capabilities; (5) the absence of databases that could link design and processing parameters with test results; (6) inadequate and expensive MMIC packaging; and (7) the high cost of test.
To achieve these goals, challenges were overcome on a number of technical fronts. Process control monitors were introduced across all FABs participating in the MIMIC program. The adoption of PCMs required the development of rapid on-wafer DC and RF test. Statistical process control (SPC) and PCM are terms that are very familiar today, but were very seldom present in the thinking in GaAs FABs of 1980 or 1985. Charts were generated, more to show that the number of wafers processed were in the 10s or 100s, rather than applying the results for process improvement or control. The ability to generate test data outstripped the engineers’ ability to read the charts, understand their significance and take specific action. First, clear correlations had to be established though they were well hidden by variations in process, test and design. Ultimately, characterization of every wafer run resulted in several benefits: (1) FAB controls based on SPC could be put in place leading to more consistent process performance; (2) correlations between DC/RF parameters and process parameters could be established based on meaningful statistics; (3) statistically valid device models could be developed and used as the basis for MMIC CAD systems; (4) first-pass design success became a realistic possibility.5 The simple introduction and utilization of meaningful PCM structures, on wafer, directed the industry on a path toward increased process control, improved device understanding, higher production yields and ultimately lower production costs.
The ability to test full MMICs on wafer was a key element in the learning process. One of the most challenging areas was the development of on-wafer power amplifier testing. Under a DARPA MIMIC Phase III program, we developed a pulsed power on-wafer test (see Figure 1). This capability revolutionized power MMIC testing allowing one to gain meaningful amplifier data without going through the expense and difficulty of assembly prior to characterization. The development of this test technology was one of the last requirements to accomplish known-good-die (KGD) testing for phased-array radar T/R module assembly.
Over the course of the MIMIC program, the production cost dropped from $20/mm2 to under $10/mm2 by the end of Phase I, and under $1/mm2 by the end of Phase II. Today’s high volume commercial production of MMICs can achieve a cost on the order of $0.10/mm2. The MIMIC programs focus on manufacturing coupled with its dual-use technology policy6 placed US MMIC manufacturers in a dominant market position. The program’s legacy persists today.
While the MIMIC program was essential in establishing the foundation for MMIC manufacturing, volume production truly drove the technology forward. The first high volume insertions of MMIC technology were for defense applications. HARM (High Speed Anti-Radiation Missile) and COBRA (Counter Battery Radar) were two of the first MMIC insertions that benefited from the MIMIC program. In the case of COBRA, the system was a C-band phased-array radar that was initially developed by General Electric Electronics Lab in Syracuse, NY, and later produced in Moorestown, NJ. The T/R module contained six GaAs-based MMICs: a driver, two combined high power amplifiers, a phase shifter, a variable gain amplifier and a low noise amplifier. The program required delivery of 25,000 chip sets to very strict specifications. Given the maturity of the technology, the use of known-good-die in module assembly was essential to achieve practical yields at the module assembly level. Production drove learning. Issues from MMIC design, to fabrication, to test, to burn-in, to assembly all benefited from the demands of production.7
Visual inspection raised significant issues to overcome. MIL standards, derived from the silicon industry, were being applied. There was no statistically validated correlation between visual inspection defects and reliability. At 1000X, a one micron or less gate, covered by two layers of silicon nitride, is difficult to see on an optical microscope. Inspection yields were operator dependant and decreased as a function of repeat inspections. No correlation was found between many of the visual yield defects and yields at burn-in and/or life test. Production MMICs were subjected to extensive reliability testing in an effort to seek correlations between process and inspection data and long-term device performance (see Figure 2). One hundred percent electrical test was also a critical area of focus. To assure high module yield, the MMICs were mounted on carriers which included decoupling networks, bias networks and stabilization networks to facilitate 100 percent RF test and, in the case of the PA and driver, burn-in. RF test was achieved by fixturing the MMICs, on carrier, in a multi-up format (see Figures 3 and 4). The use of the multi-up format allowed for highly automated testing of a large number of HPA assemblies. The die were then probed using the testing capability established in MIMIC Phase 2 and Phase 3 (pulsed power) (see Figure 5). The use of on-wafer test equipment allowed for rapid, automated test of multiple parts. This known-good-die approach was credited as one of the major factors in the success of the program.
Most foundries were pursuing a dual-use strategy to fill their GaAs capacity. While defense insertions were critical in driving both the technology and the manufacturing capability, true high volume was to be found in the commercial market place. The first major volume drive from the commercial arena came from the emerging wireless communications market. GaAs FET MMIC switches, a relatively simple product, proved to be an ideal solution for the 900 MHz wireless handset. These switches were available in SOIC-8 packages and met the performance and cost expectations for the market. This market opportunity drove the number of MMICs delivered per year from tens of thousands to millions.8 Moving from the military market, in which state-of-the-art device performance and characterization is paramount, to the commercial market, in which predictable, repeatable performance is considered a given and cost is the primary product differentiator, required a new level of manufacturing discipline. It was found that product test was a major cost driver.9 Driving down the cost of test required a more sophisticated understanding of device performance, allowing correlations to be established between various device parameters, enabling a reduction in the number of parameters tested to assure performance. A focus on hardware and software was further required to reduce the test time. Gravity fed auto-handlers coupled with robust test interface boards were implemented to achieve rapid part insertion and accurate measurement (see Figure 6). Efficient flow of data to and from test systems was also critical to improving throughput. Test times were reduced from 45 to 60 seconds per device to close to one second per device. These improvements drove down test cost and enabled high volume production.
In addition to GaAs MMIC switches for handset applications, the wireless market drove a need for high power switches for base stations (see Figure 7). Initial products, assembled in traditional metal and ceramic packages, were not amenable to low cost, high volume assembly and test. Evolution of base station technology and the demand for more complex switching functions drove more sophisticated solutions. Multi-chip modules for switch matrices drove the adoption of new assembly and packaging technology leading to cost-effective solutions. Figure 8 shows a multi-chip 4 × 6 switch matrix module containing six single-pole four-throw switches, switch drivers and power dividers, encapsulated in epoxy.
The production capability demonstrated on switches for wireless base stations and handsets was used as a springboard to begin to address multifunction ICs in high volume. For defense applications, significant integration in GaAs had been demonstrated (see Figure 9) in the form of complete transmit/receive functions for C-band radar on a single chip. Commercial pull came from the evolution of the wireless market. A high performance RFIC chip set was developed for the Japanese Handy Phone System (PHS) at 1.9 GHz.10 This work represented the most highly integrated chip set for the PHS application allowing the RF portion of the phone to be realized with less board space facilitating a smaller phone size. It became clear that to enter the high volume PHS market, the GaAs transceiver must be sold for less than $4. The only way this cost target could be met was to focus on minimizing the total area of the die. The need for compaction drove more efforts in electromagnetic (EM) simulation and tricks to reduce coupling between adjacent design elements. The transceiver IC was 3.5 mm2, realized in an ion-implanted E/D process, encompassing Rx LNA, Rx mixer, LO switch, LO amp, Tx amplifier mixer, Tx amplifier and step attenuator functions. The output IC consisted of a driver amplifier, power amplifier and T/R switch realized in 1.5 mm2. Similar parts, designed by a team of leading experts in the field but with a defense orientation, achieved similar results in 9 mm2 of GaAs, a result which did not match the target market. The two-chip solution was plastic packaged in shrink small outline packages (SSOP) allowing standard pick and place machines to be used in mass production. The packaged parts were assembled with standard wire bonding and transfer molding techniques with a total cost of tens of cents. The use of standard plastic packaging enabled the adoption of mainstream, silicon IC, auto-handlers to be used to perform functional RF testing.
The total volume of GaAs RFICs had grown to a point that dedicated RF handlers and test systems were becoming available. Further evolution of the wireless communications market led to highly integrated MMIC up-converters/drivers for cellular and PCS CDMA handsets.11 This solution consisted of two multi-function ICs that operate in the 800 MHz (cellular) and 1900 MHz (PCS) frequency bands. Volume continued to grow as the wireless industry expanded.
During this time period, the market was also beginning to accept the GaAs heterojunction bipolar transistor (HBT) as a viable solution for the output power amplifier for handheld wireless applications. Offering distinct advantages over silicon in efficiency, gain and linearity, HBTs have become the technology of choice for this function. GaAs-based D-mode MESFETs and pHEMTs, both contenders of the PA socket, lost market share due to the negative gate voltage, which required a DC-DC converter, thus increasing cost. While E-mode devices were a possible solution, they were generally found to be difficult to manufacture. Wireless communications, as the largest consumer electronics market,12 has driven volume applications for GaAs-based RFICs. This commercial pull resulted in improvement in processes, design, test and packaging.
Integration, Scaling and the Impact of Silicon
The drive to reduce cost and increase functionality mandates greater and greater integration. In the case of the wireless handset, front-end modules (FEM) have been adopted as an integration path. Clearly, no advantage is gained from merely moving components from “on the board” to inside the FEM package.13 What is limiting integration, in a cost-effective manner, is the large number of components realized in disparate technologies. Alternate technologies must be adopted for some of the key functions such as filtering and passive components.
Scaling, a driver of the silicon industry for over 30 years, does not really exist in RF semiconductor technologies. Although gate lengths are reduced to achieve higher frequency performance, the channel-to-channel pitch does not reduce much. This is due to thermal dissipation considerations as much as the fabrication tool set being used. Power devices are limited more by the substrate thickness then by the critical dimension of the transistor. The material set used in fabrication is essentially fixed: GaAs substrate, Si3N4 dielectric and gold interconnects; therefore, inductor and capacitor sizes remain fairly constant. Three-dimensional integration of passive components through the use of multiple metal layers is a possibility; implementation of this concept in GaAs has been slow due to the limited numbers of metal layers and the absences of planarization as compared to silicon processing.
The capability of adding SiGe epitaxial layers to a standard CMOS or BiCMOS process so that high speed HBTs may be integrated with conventional Si circuits has revolutionized the course of microwave circuit design over the past few years. The ability to have denser functionality, and better control over system partitioning between RF and digital domains, coupled with the economies of scale and portability that conventional silicon fabrication offers, makes extending the design of both analog and mixed-signal ICs to microwave and millimeter-wave frequencies an obvious area to exploit. SiGe can offer the opportunity to provide very low cost microwave and millimeter-wave solutions with the potential to integrate digital and control functions together with RF. This results in great flexibility in system design and partitioning.
SiGe is not the magic bullet for RF semiconductors; there are a number of performance limitations where it is inferior to traditional III-V semiconductors. In particular, power output and noise figure are two critical areas of concern. Silicon-based technologies do offer a significant benefit in terms of integration and cost in applications in which the technical performance is appropriate.
System-on-a-chip (SoC) solutions based on silicon technologies are emerging for many RF applications. At lower frequencies and lower power levels, integration on silicon is a natural approach. For example, SoC solutions for Bluetooth14 have been introduced which realize the full RF front end, baseband processor, micro-processor, memory and I/O functions in 0.25 ?m silicon CMOS technology. Integration is resulting in silicon solutions consuming more and more of the functions that had been the domain of GaAs. All RF functions with the exception of the switch and the PA, the main components of the FEM, are being integrated into silicon.
As silicon CMOS technology continues to scale to shorter and shorter critical dimensions, higher frequency applications become within the reach of mainstream silicon processing technology.15 Single chip solutions for WLAN and HIPERLAN systems are available. Through the use of SiGe technology, the millimeter-wave regime has become accessible to silicon technologies. Short-range, ultrawideband radar sensors for the automotive market are an excellent example of the potential for SiGe technology to facilitate the integration of millimeter-wave and digital functions to create a highly integrated, compact, cost-effective solution for a very high volume commercial application.16 The short-range radar operates in the 22 to 29 GHz band, a domain typically addressed with III-V solutions. Realized as a transmit IC and a receive IC, the SiGe solution in low cost QFN (quad flat-pack no-leads) plastic packaging provide full RF front-end functionality of the radar system. The receiver chip includes two LNAs, a switch, two mixers, two variable gain amplifiers and two integrators. The receiver has 45 dB (maximum) of conversion gain with 7.8 dB (including the plastic package) noise figure at 24 GHz.
Implications for the Future
The field of RF semiconductors is flourishing. The major commercial market driver, wireless communications, has become ubiquitous. The industry is expecting to pass the landmark of one billion handsets manufactured in one year. We communicate wirelessly, our computers communicate wirelessly and soon many of the appliances within our homes will communicate wirelessly. Defense applications remain demanding; on one hand very large aperture phased-array radars are driving unique requirements for very low power consumption at very low cost, on the other hand the desire to look farther with greater resolution is driving a need for ever-increasing power output and bandwidth.
The pronounced difference in volume between commercial and defense markets has, to a certain extent, driven a bifurcation in the RF semiconductor industry. More and more, companies that have focused on the commercial marketplace have done so to the exclusion of traditional defense business; while many of the FABs that benefited from the early push of the MIMIC program have either exited the market or turned primarily captive in nature. Few facilities continue to aggressively pursue a dual-use product strategy. The volume disparity between those FABs focused on commercial production and those focused on defense production is very wide. The cycles of learning with commensurate improvements in performance, yield and cost will favor those addressing higher volume applications.
Silicon CMOS and SiGe technologies are now addressing many applications that were viewed as owned by GaAs only a few years ago. Advanced electromagnetic modeling, multi-layer interconnects and short gate length active devices enable digital and RF integration of an unprecedented level. The silicon content of RF devices will only grow with time while GaAs works to maintain its grip on power and high frequency applications. Total semiconductor sales in 2005 were on the order of $225 B, of which GaAs MMIC devices represented less than $3 B. Silicon processed over six billion square inches of material, while the GaAs industry consumed 10 to 20 million square inches. GaAs RF semiconductor technology is clearly a niche of the broader semiconductor market. The GaAs niche can and must look to silicon to adopt their best practices and learn.
New RF semiconductor technologies are emerging: SiC and GaN to address high power applications; antimonide-based materials to address very low voltage applications. Significant government funding has helped move these technologies forward; impressive results have been achieved. Will the development time-line mirror that of GaAs? Are there significant applications on the horizon that will drive the learning required to mature these technologies?
Delivering high volume specification-compliant and repeatable ICs on time drove the maturation of GaAs MMIC technology. Real demand drove learning and progress in all aspects of the design and manufacturing process. Each major cycle of learning resulted in a major step forward in terms of product capability and cost. The continuation of this drive will bring RF semiconductors into applications that have yet to be dreamt.
1. H. Sobol and K. Tomiyasu, “Milestones of Microwaves,” IEEE Transactions on Microwave Theory and Techniques, MTT-50, March 2002, p. 594.
2. D. McQuiddy, J. Wassel, J.B. Lagrange and W. Wisseman, “Monolithic Microwave Integrated Circuits: A Historical Perspective,” IEEE Transactions on Microwave Theory and Techniques, MTT-32, September 1984, p. 997.
3. H. Howe, “Microwave Integrated Circuits – A Historical Perspective,” IEEE Transactions on Microwave Theory and Techniques, MTT-32, September 1984, p. 991.
4. S. Roosild and A. Firstenberg, GaAs Integrated Circuits Design and Technology, J. Mun, Ed., BPS Professional Books, 1988, p. 450.
5. E. Cohen, “MIMIC from the Department of Defense Perspective,” IEEE Transactions on Microwave Theory and Techniques, MTT-38, September 1990, p. 1171.
6. W. Pittman, “Evolution of the DoD Microwave & Millimeter-wave Monolithic IC Program,” IEEE Technology and Society Magazine, Spring 2003, p. 40.
7. R. Kole and S. Ozga, “COBRA – Lessons Learned,” IEEE MTT-S Digest, 1994, p. 1423.
8. P. Bacon, C. Mahon, D. Fryklund and P. Zampardi, “Semiconductor Trends in Wireless Handsets,” Microwave Journal®, Vol. 48, No. 6, June 2005, p. 22.
9. P. Ersland, S. Cousineau, J. Mahon and J.P. Lanteri, “Manufacturing Test Technologies for Commercial RF/Microwave Integrated Circuits,” IEEE Transactions on Microwave Theory and Techniques, 1995, p. 99.
10. F. McGrath, K. Jackson, E. Heaney, A. Douglas, W. Fahey, R. Pratt and T. Begnoche, “A 1.9 GHz GaAs Chip Set for the Personal Handyphone System,” IEEE Transactions on Microwave Theory and Techniques, MTT-43, July 1995, p. 1733.
11. K. Harrington, M. McNamara, M. Murphy, B. Khabbaz, M. Hafizi, F. Ali, C. Wakeman and R. Moazzam, “Highly Integrated MMIC Up-converters/Drivers for Cellular and PCS CDMA Handsets,” IEEE.
12. J. Neal, “The RF Semiconductor Evolution: From Single Components to Systems Solutions,” Microwave Journal®, Vol. 48, No. 2, February 2005, p. 22.
13. R. Jos, “Technology Developments Driving an Evolution of Cellular Phone Power Amplifiers to Integrated RF Front-end Modules,” IEEE Journal of Solid-State Circuits, 36, September 2001, p. 1382.
14. F. Eynde, J. Schmit, V. Charlier, R. Alexandre, C. Sturman, K. Coffin, B. Mollekens, J. Craninckx, S. Terrijn, A. Monterastelli, S. Beerens, P. Goetschalckx, M. Ingels, D. Joos, S. Guncer and A. Pontioglu, “A Fully-integrated Single-chip SoC for Bluetooth,” Solid State Circuits Conference, 2001, p. 196.
15. T. Lee and S. Wong, “CMOS RF Integrated Circuits at 5 GHz and Beyond,” Proceedings of the IEEE, 88, October 2000, p. 1560.
16. I. Gresham, A. Jenkins, R. Egri, C. Eswarappa, N. Kinayman, N. Jain, R. Anderson, F. Kolak, R. Wohlert, S. Bawell, J. Bennett and J.P. Lanteri, “Ultrawideband Radar Sensors for Short-range Vehicular Applications,” IEEE Transactions on Microwave Theory and Techniques, MTT-52, September 2004, p. 2105.
Jean-Pierre Lanteri holds an MS degree in electronics and a PhD degree in semiconductor electronics. He has been with M/A-COM for over 20 years. He is presently vice president and technical director of M/A-COM’s Strategic R&D group, with primary research interests in packaging and silicon-based system-on-chip (SoC) design for MW/mmW applications, such as automotive radar transceivers, military radar transceivers, baseband power amplifier linearization and broadband antennas. His personal focus and expertise is in low cost packaging of MW/mmW T/R modules and broadband circuits in plastic leadframes or laminate BGAs. He presently oversees an AFRL panel antenna program for space applications. A few years ago he was principal investigator for a DARPA/AFRL MAFET module program with chip-on-board components. He has extensive experience in MMIC fabrication, test and assembly, having led on-wafer test activities for DARPA’s MIMIC program and pioneered on-wafer pulse power test of MMIC PAs extensively used in T/R modules. Later he established automated assembly and test facilities producing T/R subassemblies for the COBRA and GBR radars, 77 GHz cruise control sensors and large volume commercial RFICs.
Douglas Carlson received his Sc.B degree in electronic materials from Brown University in 1983 and his Sc.D degree in electronic materials from the Massachusetts Institute of Technology in 1989. He is chief technology officer for M/A-COM’s Integrated Products Business Unit. He has been employed at M/A-COM for over 15 years working in various engineering and management positions involving GaAs materials and devices. He has been involved in compound semiconductor research and production for over 20 years.