Lasers make the grade at GaAs fabs

Compound semiconductor manufacturers tend to keep their introductions of innovative processes and technologies well under wraps. After a while, and typically at the CS Mantech Conference, a few details might be revealed, but by then the process will probably have been qualified for at least a year.

At this year s Mantech, held in the north Chicago suburb of Wheeling, TriQuint Semiconductor revealed that it has been using lasers to dice its GaAs RFIC wafers for quite some time. In itself this revelation is not particularly surprising – the Oregon-based company and many of its competitors have spent the past couple of years implementing the technology in a bid to speed wafer throughput. Perhaps more of a surprise is TriQuint s acknowledgement of the problems that it encountered along the way.

Travis Abshere presented TriQuint s "lessons learned from laser dicing" paper at Mantech. Although the laser systems from Advanced Laser Separation International (ALSI) ultimately proved to be a great success, the development and qualification period had clearly not been plain sailing.

The primary motivation behind a move to laser dicing is to increase wafer throughput. The TriQuint team had calculated that a five-fold improvement over conventional diamond sawing would provide a compelling return on its investment. In addition to the faster processing, laser dicing offers a potential reduction in die size that effectively increases the capacity of each wafer, and of the wafer fab in general. Abshere described how the reduction in street widths (the gaps between adjacent devices on the wafer) from 50 to 25 µm, possible with laser equipment, would effectively yield an "extra" wafer from each batch that is processed.

Before evaluating the technique for production, Abshere and his colleagues had identified what they saw as being the most likely problems: damage to the dielectric material left in the "streets", changes in the appearance of the diced edge of the die, potential die-cracking, and so on.

Yield drop
When TriQuint introduced laser dicing it saw a drop in yields. While that may not have been a surprise, the reason was unexpected – power amplifiers based on GaAs die were failing because of shorting caused by wire bonds. As it turned out, one of TriQuint s subcontract assemblers (SCAs) was using a process that caused these wires, which provide interchip bonding, to sag and short out on exposed metal "zippers" that form the edge of the die. "With laser-diced chips, the bond wire could sag to the point that it contacted the metal zipper, and cause a short," Abshere explained.

Normally these sagging bonds would not pose any problems because the same SiN coating that protects die from scratches would also protect the metal zippers. However, in developing the laser-dicing process it was found not only that the laser could damage the SiN that coated the die streets but also that this damage could extend beyond the metal zipper protection and into the die itself. To get round this, TriQuint removed SiN from the die streets selectively. However, the opening created by this extra step extended to the center of the metal zipper, ultimately exposing the metal to the drooping wire bond.

To solve the problem, TriQuint made two significant changes. First, it told the SCA to address the sagging bonds. Second, it made sure that the layer of SiN covered the metal zippers and extended a couple of microns into the die street for good measure.

Another technical problem arose because of the narrower cutting ability of the laser system compared with the diamond saw. This narrower "kerf" means that completed wafers have to be more carefully handled to make sure that adjacent die do not rub up against each other and cause damage, and so-called "hoop rings" are needed to stretch the wafers and keep the die well separated. With diamond saws, the larger kerf means that this isn t a problem.

All three of TriQuint s SCAs had prior experience of using hoop rings, so they gave the green light to proceed. The only problem was that none of them had ever used hoop rings suitable for GaAs wafers before. So when the stretched wafers arrived, the hoops did not fit the saw frames being used and the SCAs could not pick die all of the way to the edge of the wafers.

Lesson two
"The lesson learned here is that complete samples need to be provided to SCAs sooner," said Abshere. "It turned out that none of the SCAs had actually used small rings with 6 inch wafers." The solution was simple enough: TriQuint switched to larger saw frames and hoop rings, but this increased the process cost and caused more delays.

The third problem revolved around die cracking, something that TriQuint had already flagged up as the major risk of a switch to the laser dicing approach. After introducing this new technique, the company began to see die cracks and a high failure rate on laser-diced devices. However, the problem was not quite the one that it had previously predicted.

TriQuint s original concern was that the shape of the "cuts" from the laser would cause cracks. So after consulting with ALSI and other GaAs manufacturers that were using laser dicing, it introduced an etch-cleaning step designed to strengthen the edges of the laser-processed die.

Things did not go quite according to plan, though. Although the etch clean worked on the top of the GaAs wafers, it did not strengthen the edges. The inevitable result was die cracking, and TriQuint had to stop all laser dicing until it could solve the problem.

"We had viewed the etch clean as an opportunity to avoid dealing with a protect-coat process rather than a critical change for avoiding weak die edges," Abshere admitted. After testing the die strength in collaboration with ALSI, the team concluded that the problem was largely a result of the etch clean equipment being used, and that the wafers ought to be stretched on a hoop ring prior to the etch step to ensure a good clean. Buying a tool with greater automation and using hoop rings allowed more room for the cleaning solvent to get in between individual die. This resulted in much stronger die edges, even compared with those produced using a diamond saw.

TriQuint is far from alone in its use of laser dicing. RF Micro Devices – the biggest volume manufacturer of GaAs RFICs – was probably the first to try out the technology and is thought by some in the industry to have as many as eight laser-dicing machines.

Another leading manufacturer has been using ALSI equipment in its wafer fab since summer 2007, following an extensive trial period. Like TriQuint, it is very happy with the results and intends to purchase further laser equipment for future manufacturing expansion. "We are using lasers for pretty much everything," said a source from the company, who wished to retain his anonymity and that of his employer. "To all intents and purposes it has eliminated the scribe-and-break process," he added.

This second company has not had problems with die cracking, and it is clear that, although it may have been slower on the uptake than its rivals, it has benefited from TriQuint s early-adopter experiences.

For the unnamed company, a five-fold increase in wafer throughput has meant that far fewer wafers have needed to be outsourced for processing. "Not only is laser dicing much faster; the yields are also as good as diamond saws," said the firm. For typical 6 inch wafers with an average die count, its two ALSI machines can each process around 600 wafers per week. "That is way more than is possible with a diamond saw," said the unnamed source. The real-world figure matches pretty well with ALSI s own claim that its tool processes a typical 6 inch wafer in around 10 minutes (equivalent to just over 1000 wafers per week in a fab running round the clock).

ALSI s seems to be the laser-dicing equipment of choice for GaAs fabs. Rene Hendriks, director of commerce at the Netherlands-based firm, told Compound Semiconductor: "We have installed our system at all major RFIC manufacturers in the US and Asia."

The technology was originally developed at Philips, and this pedigree has been an advantage. Users have been impressed by the company s approach – rather than adapting a conventional system by replacing the diamond saw with a high-power laser, ALSI developed an optimized laser-dicing system from scratch. Hendriks explained: "The laser process is based on many relatively low-power individual beams, which minimizes the heat impact and damage to the wafers and adds to the overall productivity."

In the end, Abshere believes that it only took TriQuint a year to see a return on its investment in laser-dicing equipment, despite the unexpected problems that were encountered. For other GaAs fab managers, who can learn from the TriQuint experience, that period should be shorter and the business case appears pretty compelling, provided that they are able to work through any problems that do arise.

Further reading
T Abshere et al. 2008 CS Mantech Digest 69.   

View pdf of article