By Keith Gurnett and Tom Adams
The attractiveness of gallium arsenide (GaAs) as a semiconductor material is based on its ability to handle high frequencies at high power levels-something that cannot be done with silicon. Silicon devices can operate in the gigahertz range, but only at very low power levels, and low power levels are unacceptable in many military and aerospace communications applications. A missile whose continuous corrective guidance depends on swift communication with a base or satellite needs both high frequency and high power transmission capability.
One of the weak points in the production of GaAs devices has been the dicing of the wafer coming out of the fab. Despite its high performance capabilities, GaAs is far more brittle than silicon. Silicon wafers can be diced by either of the two oldest methods available-scribe and break, and sawing.
Scribe and break may break an occasional wafer, and sawing may cause chipout or corner breakage on the individual die, but both methods are widely used on silicon and, in terms of equipment and consumable costs, are relatively inexpensive.
The problem with GaAs is once it begins to fracture it tends to keep on fracturing. An anomaly that would cause a routine chipout in silicon may cause far more damage in GaAs; the fabricator must be willing to accept fairly high losses. Actually, two types of losses are involved: immediate yield losses from die that are ruined during dicing, and long-term reliability failures, where a device or system fails in service because of damage during dicing.
Within the past few years, laser-based methods have to some degree begun to replace the older, mechanical methods for dicing GaAs. In some instances the choice of dicing method involves a combination of economic and geographic factors: a more expensive and more complex laser-based dicing system might be feasible to support in the U.S. or in Europe, but the limitations of long-distance support might mean that a fab located in Asia will use mechanical dicing methods. But the decision to upgrade a dicing method at any location is filled with many complexities, and many GaAs makers currently prefer not to discuss any upcoming changes in their dicing technology.
One of the complexities involves the continued evolution of semiconductor materials. Silicon was, in a sense, the “original” semiconductor material. GaAs emerged as a somewhat expensive material for chips in niche applications, and gradually evolved as a widely accepted material for high-power, high frequency communications. Gallium nitride (GaN) has only recently entered into production for communications applications.
The advantages of speed and power offered by GaAs and GaN are now being exploited by the US military in applications such as all-weather radar, surveillance, electronic attack, reconnaissance, and communication systems. GaN is especially important in building Power Amplifiers (PA), where GaN provides several advantages: higher operating voltages, better thermal conduction and wider transmission bandwidth made possible by much better device linearity than can be achieved with silicon.
Linearity means that each communication channel is clearly defined and has sharp edges. A direct result of better linearity is access to more channels. In the commercial world, the linearity of GaN is starting to bring about significant changes in the cell phone business. The power amplifiers that carry transmissions from the cell phone tower to the customer (or to a relay station) have conventionally been based on silicon ICs.
GaN power amplifiers operate at significantly higher power levels, have higher linearity and sharper edges, and therefore permit many more channels to be used. In commercial communications, one result is that the number of power amplifiers, or even the number of cell phone towers, can be reduced.
One area where GaN power amplifiers are needed is in the development of military RF communication and radar systems in the millimeter wavelength range, where frequencies range from 10 GHz to 100 GHz. Most of these applications rely on sophisticated antenna arrays and can be carried in air, ship and mobile platforms.
Among the significant applications is the identification of targets and controlled guidance systems. One of the most significant roles for these new high-frequency, high-power amplifiers will be in satellite surveillance, and particularly in the integration of satellite surveillance into battle plan control. Research is currently under way at several locations.
GaN can also operate in higher ambient temperatures than silicon, and in fact is most efficient at higher temperatures. Silicon normally operates at a junction temperature of 180 degrees Celsius, while GaN operates at 300 C. GaN also acquires, at least in theory, higher long-term reliability because it adheres to the rule of thumb that materials are more reliable when operated well below their manufacturing temperatures. While GaN operates most efficiently at 300 C, it is processed at temperatures above 700 C.
The higher temperatures rule out the plastic packaging that is typically used with silicon devices. GaAs and GaN devices typically require ceramic packaging, which is more expensive than plastic, but which provides the advantage that the life expectancy of devices using hermetic ceramic packaging has a fully documented history, especially in space applications.
There are advances in other areas as well. Some micro electro mechanical systems (MEMS) devices have been overcoming the power/frequency limitations of silicon by using a GaAs thin chip device inside the MEMS cavity. The GaAs chip lets designers incorporate very high-speed functions into the otherwise silicon MEMS.
While silicon wafers are often diced by sawing or by scribe-and-break, these methods are not as suitable for GaAs and for GaN. There are important differences in these two materials. A GaAs wafer is a slice of pure GaAs, cut from a GaAs ingot. The GaAs wafers are sliced approximately 625 microns thick due to the material’s fragility. Before dicing, the wafer has to be thinned down to 50 microns in order to reduce its thermal resistance, since GaAs is not a good thermal conductor.
GaN cannot be easily grown as an ingot and currently there is no production process to achieve this. It is normally a grown epitaxial layer, using metal-organic chemical vapor deposition (MOCVD), on a variety of substrates, the two most common of which are diamond and silicon carbide. GaN is grown at very high temperatures, up to 1000 C, and these substrates can withstand these temperatures.
Recent developments from BluGlass, an Australian Company spun out from Macquarie University, claim to be able to grow the GaN layer at temperatures in the region of 500 to 700 C and this makes it compatible with glass and silicon substrates. It is said that the new process avoids the use of expensive MOCVD equipment. This development could change the cost structure in the manufacture of GaN layered devices.
The GaN layer deposited onto a substrate is only 1 to 2 microns thick, explains Dr. Felix Ejeckam, CEO of Group4 Labs, a firm that uses a patented method to deposit GaN onto diamond wafers. Across the industry, he explains, there are actually 5 different substrates that are used for GaN - diamond, silicon carbide, sapphire, silicon, and a bulk (but non-crystalline) form of GaN itself.
“The brittleness of GaN completely depends on the brittleness of the underlying substrate. The thin film is only 1 or 2 microns thick. The substrate, however, is 200, 300, or 400 microns thick,” he adds. The substrate materials are all very tough, he notes, and are always diced by some form of laser; sawing and scribe-and-break won’t work here.
GaAs wafers are often diced by mechanical methods, although there is some cost in terms of breakage. Laser dicing methods, though, are being used more frequently on GaAs. The more specialized GaN applications will probably never amount to more than a small fraction of GaAs applications, making clean, low-loss dicing of GaAs wafers something of a priority. While laser-based methods avoid the mechanical problems of saws and scribe-and-break, they make it easier to create thermal stress that will lead to cracking in the GaAs wafer. Laser-based GaAs dicing tools work best if they can cut with reasonable speed and can keep the wafer relatively cool.
One way to achieve these desired functions is to split the laser into multiple beams. A dicing system recently developed by Dutch firm ALSI uses an optical grating to split the laser into anywhere from 2 to 50 beams. Each of these beams exits the optical grating at a slightly different angle. The laser moves down the street between individual die as a line of spots that are from 15 to 100 microns apart. Each spot removes some of the GaAs. After it moves on, the GaAs cools before the arrival of the next spot. As a result, the GaAs never becomes sufficiently heated to permit chip-outs or fractures. A vacuum system removes debris from the cutting process.
Before a wafer is diced by the laser, it is coated with a protective material that is removed by water after dicing. A realistic cutting speed for the laser is around 200 millimeters per second. The net cutting speed is greater if the wafer is thinner, because a very thick wafer may require more than one pass. But many GaAs wafers today have been thinned to 100 microns or even less, which means that cutting can be done in one pass.
An unusual feature of the split laser is its ability to make a very narrow cut, or kerf, in the GaAs. In wafers that have a thickness of 100 microns or less, the split laser removes material from a kerf that is only 10 microns wide-about half the width achieved by other laser systems. As a rule of thumb, the kerf width is one-tenth of the wafer thickness-e.g., 30 microns for a 300-micron thick wafer-but does not become narrower than 10 microns.
The width of the kerf is important because it determines the width of the street between the rows of die on the wafer. GaAs is widely used in making discrete semiconductor devices such as transistors and diodes, and in making blue LEDs. All of these are very small die having a width of 0.24 millimeter (240 microns). Conventionally, the width of the street might be 50 microns. Using the split beam laser lets designers reduce the street width to a little more than 10 microns. Since the streets run in both x and y axes on the wafer, the fab can now place 30% to 40% more 240-micron die on a wafer of the same size. For larger die, the gain in number of die per wafer is less, but still significant.
One type of device to benefit from this technology is GaAs monolithic microwave integrated circuits, or MMICs, which are used in satellite applications. Their frequencies range up to 10 gigahertz, a level that is not easily achieved in silicon. The MMIC wafers have a gold metallization eight to nine microns thick on the back side that is used for die bonding in later assembly processes. At these power levels, die cannot be adhesively bonded because the thermal extraction requirements are too high. An alloying process establishes a good thermal path to the package and the outside environment. The presence of the ductile gold means that the wafers cannot be sawed; the gold will smear onto the saw blade and quickly put it out of operation.
