FPGAs are a multibillion-dollar business, characterized by suppliers like Xilinx, MicroChip, Texas Instruments and Cypress Semiconductor. So why are they reliant on a single source for copper-wrapped solder column attachment?
And not just any source. Six Sigma is a small, founder-run Silicon Valley-based company. For more than 30 years, it has performed a variety of third-party services, including solder dipping, BGA reballing, and solderability testing. It also happens to be the sole supplier of copper-wrapped solder column attachment services to the major FPGA vendors. And according to industry watchers, that leaves the supply chain in something of a pickle.
The risk with any sole source is something happens that affects their ability to make deliveries. Such an outcome would spell disaster for the high-rel companies that use column grid arrays (CGAs). And loss of access to the unique copper-wrapped columns used in key programs, including military and space, would put those applications at severe risk.
One alternative is organic packages. An investigation published by Reza Ghaffarian at Jet Propulsion Laboratory in 2006 found solder joint reliability of plastic BGA packages on polymeric boards to be generally superior to that of the ceramic versions due to the CTE mismatch between the ceramic balls and the substrate. In 2015, new work performed by Martin Hart of TopLine found solder columns on ceramic packages to be more compliant than solder balls, and able to absorb CTE mismatch of up to 10ppm/oC when soldered on a PCB.
In his paper, Hart also asserted that a copper-wrapped version is more reliable than the conventional tin-lead version. Raychem in 1983 found it could wrap copper ribbon around high-temperature PbSn20 solder wire, then apply a eutectic SnPb coating to secure the copper to the column. These ceramic CGA (or CCGA) packages were found to outlast conventional PbSn10 columns under temperature cycling tests, Hart wrote. In 2010, Kyocera-run finite element modeling tests found the effective thermal conductivity of Six Sigma’s copper-reinforced interconnect was approximately 75% higher than the equivalent IBM high-lead CGA. In a subsequent study published in 2012, JPL tested CCGA packages for 596 thermal cycles ranging from -185° to +125°C – conditions meant to replicate those that would be experienced on Mars and Jupiter’s moons – and found no catastrophic failures, although there was evidence of solder fatigue.
The research came at the same time the industry was undergoing a major supply chain shakeup. IBM, the inventor of the column grid array, sold its CGA equipment, IP, and licenses to Silicon Turnkey Systems, which in turn was acquired by Micross Components. (IBM was a main supplier to Xilinx.) A few years earlier, BAE, another supplier, switched to providing internal services only. That left Actel, which was acquired by MicroSemi in late 2010, without a supplier of solder columns. MicroSemi then outsourced its attachment services to Six Sigma. Previously, Raychem apparently didn’t see the hoped-for revenue materialize, and subsequently licensed its patents to the last remaining supplier … Six Sigma.
There’s little question additional suppliers would impact lead-times and pricing models in a positive way for customers. But the cost of capital equipment, coupled with the lengthy qualification time, seem to be roadblocks for the few companies that considered jumping in the mix.
Hart notes the frailty of the situation. “The defense industry should be concerned about its dependency on a single source vendor to attach copper-wrapped solder columns to FPGA devices as recorded in QML-38535,” he said in a press statement. “Military-grade FPGA devices used for defense applications cannot function without solder columns.”
Help may be on the way. Reports have it that as many as four companies might be inching toward qualification. Funding remains a critical issue, however.
Knowing the potential for losing a key sole source supplier, should the US government be more vested in the outcome? Perhaps, but it seems FPGAs simply don’t rank high enough on the priority list for near-term action.
Beyond the not-so-simple matter of what the FPGA market might look like in a few years, this situation raises a host of knotty questions. In no specific order, how many other critical technologies are facing unplanned extinction? How do we speed the process so these “lower priority” items get resolved? When problems like this come about, is a consortium of the major vendors in order? And should government set up a team to instruct competitors on how to develop consortia for intensive short-term programs to resolve common industry problems like this?
P.S. As always, we are grateful for your ongoing support, and wish all our readers and customers a healthy, prosperous 2020.