Two days ago I got a phone call in the middle of the night. It was coming from a friend who used to work at a company that was a client of mine. This company, which I'll leave unnamed, has a large custom designed machine (one of a kind in the world) that is used to make honeycomb material of the type used for the couling seals for the turbine blades in aircraft jet engines. My part in the machine development was to write all the control software and work onsight to help integrate said software into a working system as the machine was deployed. The machine uses a large 1400 Watt laser to spot weld the metal strips of the honeycomb material together into sheets of this material. A pair of mirrors are used to quide the aiming of the laser beam to the weld spot on the material which is in constant motion. Thus one of the mirrors actually tracks the material motion in real time so the laser stays focused in the same physical spot and can make a good spot weld. The DOS based 33 MHz 80386 computer which hosts all of my software, including the motion controls for the machine drives did not have enough bandwidth at the time to perform control of the laser mirrors directly. So a third party was contracted to build a small 68HC11 control module and some specialized hardware to implement a mirror motion profile generator and driver.

The machine was originally deployed in the 1993 / 1994 timeframe. Later in June of 1995 I was contracted to make some performance enhancements to the machine which included the necessity to modify the 68HC11 mirror controller firmware and hardware. The circuits implemented in the mirror controller were tight and I had to add some logic to the board. To do so in the framework of the existing circuit board I re-worked in a Cypress GAL of part number PALCE16V8.

The phone call comes in the night of September 28, 2004 from the friend of mine that used to work with this machine at the company. They had called him because the machine had died. Engineers and technicians at the company had traced the failure back to the GAL on the circuit board of the 68HC11 mirror controller. He called me because they needed to find a replacement GAL. For reasons unknown to me the engineers at the company could apparently not find spare parts made in 1995 nor could they find the .JED programming file for the GAL.

As luck would have it I still had the machine control software, 68HC11 firmware and GAL design information on my computer hard drive. So of course I immediately agreed to make them some new GALs. I looked in my parts stores and found the original tube of PALCE16V8 GALs that I purchased in 1995 so I could make new parts for the customer. Now for the fun part. As I pulled fresh (but old) parts out of the tube and put them into my Bytek programmer for programming they did not pass the original blank checks. Futher investigation and I found only two parts of some 15 or so in the tube would pass a blank check. Note that back in 1995 new parts from a tube would have passed the blank check 100%. I found that by reading these parts into the programmer memory and then looking at the pattern that it appears totally random in each part. I was able to erase these parts and get them to take the new programming and verify to the desired pattern.

The next morning as I was talking directly to the machine manager at the company, (I was getting the address and account number so I could FEDEX them the replacement parts), which is located 2000+ miles from where I now live, he told me that the machine failure had been a gradual one over the period of a month or so and that the operators had suspected the 68HC11 circuit board. They had discovered that freeze spray would get the circuit board to run the mirrors when the board stopped working. So I now wonder if GAL parts in-circuit lose their stored bits as readily as erased parts in a tube in my storage area!! So much for non-volatile eeprom cell long term data storage.

Michael Karas