In aerospace and defense, there’s a lot to be said for wanting the latest and greatest. Advanced materials and miniaturization let engineers push system design into heretofore-uncharted territory.
Sometimes, though, you want the tried and true—particularly for applications in space where a failure can mean the loss of a billion-dollar satellite. In many instances, an integrated circuit based on an older design is more valuable to satellite designers than a newer, more capable one because legacy chips are already radiation-hardened and qualified for space flight while newer ones must go through a long, costly validation phase before they can be qualified.
So, imagine the consternation among satellite and space instrument designers several years ago when Actel Corp. discontinued production of a pair of field-programmable gate arrays (FPGAs) that have seen widespread acceptance in commercial and military satellites. BAE Systems was the original manufacturer of the FPGAs, with the technology and design owned by Actel.
En masse, all the major satellite manufacturers such as Boeing, Lockheed Martin, and Space Systems/Loral let BAE know that they’d have a ready market if they were willing to re-license the FPGA from Actel and restart the production line.
“It was a pure business case; there was an established customer base and we had a good idea of the quantities needed over the next few years,” said Donald Francis, Program Manager for Advanced Products at the BAE Systems Electronics & Integrated Solutions foundry in Manassas, VA.
BAE expects to manufacture between 5000 and 10,000 of the FPGAs over the next five years, with a value of $25 million to $50 million.
FPGAs are valuable to designers and engineers because they can be customized in the field and tailored to any number of applications, as opposed to programmable read-only memory (PROM) chips that are programmed at the factory.
These particular FPGAs—model RH1280 with 8000 gates and model RH1020 with 2000 gates—are typically used in communications satellites for processor control and data handling for navigation purposes.
“The reason the need is so great for older FPGAs is that government systems in general are fairly conservative systems,” said Francis. “Once they’ve been successful, there is a tendency to use the same systems if possible.
“There are a lot of satellites still making use of 10-year-old technology because if it’s qualified then manufacturers like to use it. There’s a very high cost to certify something for space, and if you can use the same type of hardware, you can save an enormous amount of money.”
BAE’s resumption of FPGA manufacturing was not just a matter of flipping a switch, however. Three years had passed between the time that production stopped and restart began, a period of time in which the foundry was completely shut down for a $100 million upgrade that was partially government funded. (It was the shut down of the Manassas foundry that helped prompt Actel to discontinue the RH1280 and RH1020 FPGAs in the first place.)
The foundry was now populated with the most advanced manufacturing equipment in the world, but had to dial down that capability to build FPGAs based on 1998 designs.
“We effectively had to build these parts identical to how they looked in the past,” said Francis. “We have a lot more capability now but couldn’t use it. The biggest challenge was building a new part with new tools and making it appear like it was built with older technology.”
The first fabrications of the FPGAs began in the fourth quarter of 2007, but not without some false starts.
“It didn’t go exactly as we expected it to,” said Francis. “We had to compensate for some tools that no longer exist in the industry. The earlier tools used temperature and pressure to implement a process. The new tools can’t modify temperature, only pressure and time.
“It’s not so much that the tools aren’t as versatile; it is more about the processes developed with the tools you had at that time. You have to do things differently to mimic the same processes.”
Fortunately for BAE, the Manassas foundry still employed 80% of the engineers who worked on the FPGAs in the 1990s. Francis credits their experience in processes and test engineering with helping the facility master the challenge of making what was old new again.
There are only a handful of foundries in the U.S. capable of making rad-hard chips. Honeywell has a facility in Plymouth, MN, while National Semiconductor and IBM devote a small part of their business to building rad-hard chips.
With so few foundries manufacturing rad-hard chips, the problems of availability and obsolescence will only get worse. That may prompt companies such as BAE to produce more of the hard-to-get components internally.
“Over the next four or five years, I expect that BAE will look hard at the parts we need for our higher assemblies and may bring some of these parts in our foundry ourselves,” said Francis.