|GaAs MMICs - Moore's Law in RF?|
|Originally published July, 2000|
|¿ 2000, 2005 Carlo Kopp|
Without doubt the most important development in microwave technology during the last decade has been the Monolithic Microwave Integrated Circuit (MMIC). The MMIC is to the RF engineer what the integrated logic gate was to the computer engineer, decades ago - the means of building complex designs with economical prepackaged functional blocks. In this month's feature we will take a closer look at the MMIC and its implications for microwave and digital communications over the coming years.
RF Design Issues
Designing analogue hardware to operate at radio frequencies (RF) has traditionally been a relatively messy process, in comparison with digital logic design, and remains very much the domain of a very specialised community of engineers. RF hardware is by its basic nature analogue, and few opportunities existed in the past practice of the design process to abstract functional blocks and avoid the frequently pathological behaviour of component level designs.
Let us consider a sixties or seventies RF design such as a low noise preamplifier used in a receiver. The available component base comprised primarily Silicon and Germanium bipolar transistors, axial lead resistors and capacitors, and ferrite and air core inductors.
Each and every one of these components exhibits parasitic or stray capacitance, inductance and in the transistors, internal capacitances resulting from the geometry of the transistor die, and the behaviour of the semiconductor junctions.
A design engineer would therefore construct a fairly elaborate mathematical model for the intended circuit's behaviour in the RF domain, and concurrently would also have to do the same for the direct current (DC) behaviour of the circuit to ensure that the transistor was operating in the intended regime and was stable with varying temperature. To complicate things, a noise performance model was also required, to account for the thermal noise in resistors and shot noise in the transistor. Of course, calculations or simulations would also be required to account for component tolerances, since RF transistor specs would be very loose, and 1% accurate resistors rather expensive.
Through repeated iterations using a calculator, or if paid for by a serious employer, a software simulator like SPICE, the engineer would zero in on the intended combination of component values to get the right balance between gain performance, noise performance and thermal stability, while looking over his shoulder for the company accountant, heard mostly complaining about project delays and extravagant choices in transistors and capacitors.
Once the theoretical design was completed, a prototype would be fabricated on a printed circuit board and tested. No chance of delegating the printed circuit board layout to a draftsman since the oddities of RF require that the primary designer produce the layout him- or herself.
Usually the prototype would oscillate due to feedback coupling at radio frequencies, and would require revisions to the board layout and judicious adjustments in component values, mostly determined empirically yet again through multiple iterations. Finally a working prototype existed.
The next phase in the process was a demonstration production unit, built to the documentation package in a production environment and carefully tested to verify its performance. More than often further iterations would follow.
Finally, the design would be transfered to production, and would usually require manual adjustment by a technician to exactly meet the intended specification. Very careful and accurate assembly was mandatory, to ensure that parasitic inductances and capacitances were not inadvertently added into the design by cutting component leads to the wrong length, or putting a bend in the wrong spot.
Was this messy ? Absolutely !
Classical RF design and manufacture was time consuming in every respect, during every portion of the process from idea to end product. Increasing the operating frequency from the HF, VHF to UHF was a headache, and going microwave a nightmare. At microwave frequencies, even tiny stray capacitances of nanoFarads and inductances of nanoHenries could wreak havoc with a design. Moreover, shielding covers or cavities to contain the circuit would contribute.
Perhaps the biggest headache of all was reproducibility in designs, since the combination of sloppy component tolerances and mechanical assembly left many opportunities for designs to deviate from the intended specification.
By the seventies RF designers with the budget to do so shifted from printed circuits to hybrid circuits, using ceramic substrates with resistors, capacitors and conductors fired on to the surface of the ceramic substrate, and active components such as transistors and diodes then soldered on to the substrate.
"Hybrids" proved to be excellent, since they allowed much more compact physical designs, and tighter production tolerances. Cost however remained a major issue, as a result of which hybrids became a staple item in equipment like radars but continued to be a cost problem for commercial and consumer equipment.
Another issue which proved to be an ongoing problem was the performance of transistors with increasing frequencies. Silicon bipolar transistors, the workhorse of the digital logic base, would suffer worsening gain problems at several GigaHertz and were frequently also noisy. While speed could be improved to a large degree by shrinking the size of the transistor, a more subtle problem arose, which was inherent in the Silicon material itself - poor electron mobility.
Mobility is a measure of how quickly charge carriers (electrons, holes) can travel in the crystalline lattice when an electric field is applied. The lower the mobility, the stronger the field required to make them move quickly. In a transistor which is trying to amplify a signal at many GigaHertz, poor mobility tranlstaes into poor gain, and gain is the measure of a transistor's worth in most of its uses.
The answer was to be found in GaAs semiconductors, rather than Silicon. GaAs has typically around six times the electron mobility of Silicon, providing the potential for significantly faster transistors. GaAs also proved to be better from a noise performance perspective, so the two key problems in an RF transistor, speed and noisiness, were ostensibly solved by the GaAs transistor.
The reality was not as tidy as was initially expected. It has taken almost two decades for GaAs components to transition from the early production components to today's mature volume products. During my undergraduate years, the standard joke in the EE community was "GaAs - the material of the future - still ...".
The problems with GaAs were manifold. The material proved to be very difficult to fabricate, the transistors proved to be very fragile and easily damaged by electrostatic discharge, overheating or electrical overload, much more so than Silicon. In the employ of one company, I was banned from using a GaAs transistor since it was expected that the production workers could not solder them in without damaging them !
The commercial pressures for more bandwidth grew very rapidly during the early nineties, with the massive growth in mobile telephone use, and the growth in the Internet. Mobile telephony proved to the key volume driver for commercial commodity GaAs components.
However, the technology base for RF could not move ahead while remaining shackled to Silicon integrated circuits and discrete GaAs transistors.
Silicon fabrication techniques allow for digital components which can be clocked well beyond a GigaHertz, and Silicon allows for tremendous density. However, combining density with low noise Silicon bipolar transistors has proven to be difficult, indeed the fastest low noise Si bipolar discretes this writier has seen are only useful to several GigaHertz.
Clearly the answer to this problem was to integrate many GaAs transistors on to a single chip, in the manner performed with Silicon successfully over the last 4 decades. This proved to be no mean feat, given the finicky nature of GaAs as a material. The Microwave Monolithic Integrated Circuit (MMIC - pronounced "mimic") proved to be an elusive goal.
A lot of expensive research was required to push the GaAs transistor from the domain of discretes into the integrated circuit (IC).
Much of the early funding and early production of GaAs ICs was paid for by the US DoD. They had a very strongly vested interest in this respect, since radar remains the key military sensor used for finding things to blow up. Whether we are building radars, or building equipment to jam radars, we require high density, reliable, economical RF building blocks. The particular prize in the military game was a device called an Active Electronically Steered Array (AESA), also known as a "phased array". An AESA is a flat panel microwave antenna which can be pointed by individually manipulating the phase shifts, or delays, of the hundreds or thousands of individual receive and transmit elements which make up the array.
With no moving parts the AESA is very reliable, can point its beam in milliseconds, yet it can be easily buried into the flat surfaces of a stealth fighter or bomber, and can be built with sidelobes 1/100 - 1/1000 the magnitude of a conventional mechanical antenna. The AESA was the radar designer's dream.
The snag with the AESA is that it needs at least 1500 and more typically 2000 individual transmit receive modules, each of which has to contain a transmitter, receiver and phase shifter, as well as the radiating element, digital control bus and RF feed connections. Operating at 10-20 GHz, each much be less than a centimetre in cross section.
Needless to say the only technology which could possibly allow the manufacture of the densely packed AESA TR modules was the GaAs MMIC. The US DoD was the first major player, but quickly followed into the fray by the EU and the Israelis, as well as the Japanese. All funded research and pilot production, and now all are either paying for the manufacture or the impending manufacture of AESAs.
Once the expensive research was done and the production techniques were refined, the manufacturers quickly turned to commercial applications, for which GaAs opened up huge possibilities: mobile telephony, satellite telephony, mobile networking, multipoint distribution, satellite communications. Any application which could benefit from a RF chip using GaAs was a candidate.
The big attraction commercially lies not only in performance, but also board level manufacturing, since economical high volume robotic component placement can be used, and many of the production tolerance problems seen with manual assembly simply go away.
The Silicon monolithic integrated circuit appeared during the
sixties and has since then revolutionised the computer industry, as well
as the consumer electronics industry. While the GaAs MMIC is a late
arrival, it promises similar revolutionary changes in RF technologies,
and many cabled high speed digital communications technologies. Silicon
will remain competitive in many lower speed RF applications, but the
high ground has now been taken by GaAs.
(Images - Siemens)
The basic building block in any solid state integrated circuit is the humble transistor, and this is no less true for a GaAs MMIC. The transistor is used as a switch, an amplifier, or if contorted in the right manner, a current source or load resistor.
In Silicon based technologies, the two "workhorse" transistor types are the classical bipolar device, and the MOSFET, a mainstay of high density digital circuits. Neither proved to be practical for GaAs.
The first GaAs transistor to achieve high production volumes, as individual disrete transistors, was the MESFET (Metal Semiconductor Field Effect Transistor).
The MESFET is a close cousin to the Silicon MOSFET, and like the MOSFET, is constructed with a source, gate, and drain (See Figure 1. by Litton). A voltage change between the transistor's gate and source pins causes a change in the current flow between the source and drain pins. Unlike the MOSFET, where the gate is insulated from the semiconductor substrate by an oxide layer, the MESFET uses a Schottky metal junction produced by applying the metal gate electrode to the semiconductor directly.
Unlike Silicon, where a MOSFET can be easily fabricated by doping a source and drain, laying down an oxide for the gate, and then putting down Aluminium connections for tracks, sources, drains and gate electrodes, GaAs MESFETs are much more demanding to build.
The MESFET is fabricated on a GaAs substrate, using Molecular Beam Epitaxy (MBE) techniques to grow extremely thin layers of doped GaAs with precise thicknesses and compositions. The bulk of the MESFET comprises a lightly negatively (N+) doped layer, over which a more heavily doped layer is placed to form a base for the source and drain contacts. Channels are etched into the substrate for the gates, which are then applied as a layered Pt/Ti/Au (Platinum, Titanium, Gold) structure and then carved to an exact trapezoidal shape using an electron beam. The gate is about half a micron in length.
The source and drain contacts are then produced using layered structures of Germanium gold alloy (GeAu), nickel and gold (GeAu/Ni/Au), the Germanium doping the underlying GaAs to improve contact performance.
While the MESFET provided a robust basic device for many applications, especially those involving low noise high speed receivers and buffers, its performance ceased to be competitive with the advent of more complex High Electron Mobility (HEMT) transistors. MESFETs do remain widely used, and are especially common in applications such as switches and attenuators.
The HEMT (Figure 2. - Litton) transistor family spans a range of devices, with manufacturers frequently using variations on the nomenclature to label their proprietary flavour of the device.
Where HEMTs differ from older MESFET devices is in the use of heterostructures, in which two different semiconductors are used to form the transistor. While any meaningful discussion of the solid state physics of GaAs HEMTs would exceed the scope of this treatment, suffice to say that the heterojunction layer between the AlGaAs and InGaAs creates conditions in which an electron gas with very high mobility is formed. As a result the transistor is significantly faster than the classical MESFET, which relies on the mobility performance of the base material alone.
Pseudomorphic HEMTs (PHEMTS) are available with useful
performance out to many tens of GigaHertz.
(Images - Siemens)
The step from the fabrication of individual transistors to complete monolithic circuits with tens to hundreds of components on a single slab of GaAs has its complexities. While the techniques for processing the materials are essentially the same, quality demands do go up since losing one transistor out of a hundred due to a fabrication defect amounts to losing a whole chip die if you are fabricating a a wafer of MMICs.
The other significant issue with MMICs is the development of design rules for the layout of components and connections, and the development of robust passive components such as PIN diodes, resistors, capacitors and inductors which may be integrated into the same slab of GaAs. Since the MMIC is a complete microwave RF circuit on a chip, the design must reflect established RF design rules.
The density of a GaAs (or Si) MMIC is much lower than that of competitive Silicon digital ICs, even if similar transistor sizes are used, simply because of the need to observe the same RF design rules which plagued RF circuit designers in decades past. Every track is a waveguide ....
An example of a 60 GHz satcom power amplifier MMIC with a 550 mW output rating (Lockheed Martin, Chao et al.) is depicted in Figure 3, and a GSM MMIC (Siemens, Kapusta) is depicted in Figure 4. Lower operating frequencies do allow some increases in density, but millimetric band operation usually allows no compromises.
The great enabler for MMIC design has been the availability of cheap compute power and excellent simulation software, which allows engineers to devise circuits using previously tested structures and frequently also building blocks.
A very wide range of GaAs MMICs are now available in the commercial marketplace, many of which are general purpose building blocks, some of which are versions of radar and AESA components, and increasingly, custom devices for established commercial applications. These devices can be supplied as dies alone for use in multichip modules or hybrids, or in conventional resin TSSOP or SSOP28 packages for robotic surface mount on printed circuit boards.
For a product designer working at the board level, a suitable range of of-the-shelf MMICs allows the rapid design and development of RF equipment with a minimum of fuss, since the most difficult bits are hidden away in the MMIC. What we are seeing today in RF design is what happened in logic design two decades ago, when MSI chips largely changed the game.
Clearly the market for GaAs MMICs is booming, between 1997 and 1998 a 200-300% growth in deliveries was observed. Manufacturers are now aiming at maximising volumes and minimising costs, and have shifted from 4 inch wafers to 6 inch wafers. Cost still remains an issue in comparison with Silicon, since the equipment for molecular beam epitaxy and electron beam shaping is extremely expensive, and the cost of the raw materials is also much higher. Typically the cost of a 4" GaAs MMIC wafer is simlar to that of a 6" Si MMIC wafer.
Market projections for this year indicate that the primary uses of GaAs MMICs, in descending order of volume, will be TV / Cable TV tuners/modems, and mobile telephones, both with about 45% of the total market volume, followed by wireless LANs with about 5% of market volume, and automotive distance warning radars with about 0.5% of market volume. Military radar and satellite comms, which were the original targets of the basic research effort, amount to about 0.25% of the market collectively.
If current trends continue, we can expect to see an ongoing decline in costs and broadening uses of GaAs MMIC technology in consumer and commercial applications. The technology will contribute to further growth in other areas, such as high speed networking and wireless networking.
Like the Internet, the GaAs MMIC was born out of the military research base yet is likely to produce its greatest impact in commercial and consumer markets. An interesting point to ponder!
|$Revision: 1.1 $|
|Last Updated: Sun Apr 24 11:22:45 GMT 2005|
|Artwork and text ¿ 2005 Carlo Kopp|