Every part that Tekmos manufacturers, has to be tested. The wafer manufacturing process will naturally produce a percentage of non-functional parts, and we must identify them and reject them. Tekmos uses a combination of inside and outside test resources to test the parts. Tekmos has experienced sizable growth over the last two years, and as a result, we were finding ourselves limited by our test capacity. To address this, we decided to buy additional testers to augment the two testers we already had.
Purchasing a tester is a complicated decision. You need to determine how many pins the tester will need. Most of our devices have less than 100 pins, though some have 356 pins. Since tester cost is generally a linear function of pin count, it is important to buy the right amount of pin capacity. You must also consider the economic tradeoffs of tester speed and vector memory capacity.
After months of internal deliberations, we decided to purchase a new Griffin III tester from HiLevel. This tester has 128 pins with a 64M vector depth and a 100 MHz clock rate. This is a significant increase in our capability, since our older testers were limited to 40 MHz and 8M of vector depth.
While we were negotiating over the Griffin II tester, we found another HiLevel on the surplus market. This one was a model ETS780 tester, with 256 pins and an 8M vector depth. It is a much older tester, but was a valuable at a reasonable price, so we acquired it as well.
Along with the testers, we have acquired a new problem. Each tester has a different mechanical interface. In the past, we have just built a custom interface for a specific part to connect to a specific tester. However, this approach has caused capacity problems as our volume has increased. We frequently find ourselves in a situation where we need two different parts that are only testable on a single tester. This results in the situation where one tester is running two shifts, while an adjacent tester is standing unused.
Our solution to this is to create an interface that will connect any tester to any hardware. Since many of our products are tested on a handler at the end of a cable, we are going to make that cable interface standard. One part of that standard is to assign the tester channel number to always be connected to part pin number. Channel 1 to connect to pin 1, and so forth. Normal industry practice is to assign tester channels to pins in a pattern that produces equal physical lengths. This allows the maximum timing accuracy. In our case, we do not need that level of accuracy. Most or our designs are old, and do not have critical timing specifications. It does not matter if signal A arrives 500 picoseconds before signal B.
Converting to the new system will be a slow process. We have perhaps 30 different boards to re-do. So we will start with the new products, and gradually extend the process to the higher volume parts. We may never convert some of our very old parts. But for the important parts, we will have flexibility. Parts from 1 to 64 pins will run everywhere. Parts from 65 to 128 pins will run on 3 testers. Parts from 129 to 256 pins will run on two testers, and parts from 257 to 320 pins will still only be supported by one tester.
Now that our tester capacity problems are addressed, we will turn our attention to our next test problem, which is adding a probe capability to our test floor.