0.8 Taking the speak n spell into production  (Page 5/5)

As I indicated earlier we did our pre-production ramp in the June/July time of 1978. That meant the design team traveled from Lubbock to Midland every week to resolve manufacturing issues. These weekly visits increased from a day or so (remember, in West Texas, 150km is only an hour or so drive), to driving out early on Monday morning, staying the week and returning Friday evening. The weekly trek to Midland lasted until Thanksgiving or early December. Christmas was coming and it might not be possible to get the product to the stores in time. We were all very thankful that Christmas.

One of the production issues kept me in Lubbock to resolve it. Due to its severity, I lived through one of those novel 3 day weeks. I started on Monday morning and worked through the night until Tuesday morning. I came back in Wednesday morning and worked through the night until Thursday morning. With the fix determined, I was asked to drive to Midland on Friday to install it. After I explained that I would not be able to stay awake driving my car to Midland and back, it was decided that a charter plane would fly me to Midland and back. As the manufacturing facility was at the end of the runway at the airport. The plane landed, taxied to the front door of the manufacturing plant for me to get off. After I had implemented the change, I called the pilot and he taxied back over to the plant. I got on and flew back to Lubbock, that was quite an experience for a young engineer.

Probably the most difficult problem we had was with the controller chip, the TMC0271. It magically became the cause of the Speak N Spell to fall out of production. This problem required people in Midland, Lubbock and Houston to work together for a resolution. The problem ended up being a manufacturing problem with the TMC0271. At the time there were two products being manufactured on the TMC0270 base set: the Speak N Spell's TMC0271 and another product, the TMC0272. For yield reasons, each of the two versions of the TMC0270 had different production flows. Over a period of several production lots the two devices swapped process flows. Once the switch was found, it was rather straight forward to get back into production.

If I didn't cover it in an earlier chapter the last hardware bug for garbled speech was finally identified. It was a clock issue. The synthesis algorithm required a 22 stage pipeline. The 22 time intervals was created by taking a base 32 counter and having it reset on the 22nd count. Unfortunately, the other 10 states in the counter were not designed to drive the count back to zero but rather started its own counter. Once out of the 22 state counter it would cause garbled speech. Fortunately long before we found the actual cause of the problem we found an un-explainable simpler fix with a few ceramic capacitors.

I'll end this chapter with one of my favorite bug fixes. The Speak N Spell was in its third or fourth year of production and had been moved from Midland to Abilene, Texas. To reduce the cost further, the potentiometer used to set up the clock frequency had been replaced with three resistors, of which one or two of the three resistors could be clipped out to bring the clock frequency into specification. As expected the Speak N Spell fell out of production as the resistor solution no longer worked on a high percentage of the devices. I was called in with my technician to help troubleshoot the problem and fix it. The product line technician had spend quite a while attempting to resolve the issue with no real success. He was not at all happy when I explained it was a date code issue. He still didn't believe it even after we ran a test where I picked the good from the bad based on the date code of the TMC0271 (again the culprit). I finally had to break down and explain that there had been a process mode change to the next smaller geometry. That meant that the simple oscillator in the device had changed characteristics and would need a new set of three resistors to get the clock frequency within the specification. I guess I should have said that earlier, but I wasn't happy being pulled back in to the product engineering role and thought I could at least have a bit of fun with it.

This last story did teach me an important lesson. It is now my "rule of the decade". It goes this way: Just because you can make one doesn't mean you can make 10, and just because you can make 10 doesn't mean you can make 100, and just because you can make 100 doesn't mean you can make 1,000, ..... and just because you can make a million doesn't mean you can make 10 million of them. Each decade of manufacturer brings a new set of production problems. Researchers believe that making one work proves it possible, designers believe that making 10 work proves it is producible, quality engineers believe that making 1,000 proves is will be reliable. It is only the product engineers, who have to stay around through the life of the product, who understand this rule.