Where there's smoke, there's a bad design

12/3/2010 8:11 AM EST

In order to alleviate the size issue, they would sometimes use a "Stroker" (as opposed to "Raster") display to simplify the circuitry and save power and cooling. A Stroker display is one that relies on directly steering the beam with the CRT deflection coils to form map outlines, letters, and numbers on the display.

This is analogous to how you would draw the letter "A" with a pencil (go ahead, try it). Where you actually draw a line consider the beam "on."Where you move a space without drawing, consider the beam "off." This was an ingenious solution to display design problems, although the display DID suffer from a pronounced flicker when the screen got too crowded.

For instance, when the CPU wanted to draw a line on a 640 X 480 display, it would execute two writes to the X and Y position registers (e.g. 100/100 and 200/200) which would in turn draw a line from position (100,100) to position (200,200) on the screen. It would have to re-execute these writes to the registers periodically to keep the line displayed on the screen. The refresh rate was dependent on the persistence of the phosphor on the screen.

A friend of mine was tasked with designing the display driver cards for the new display. The old display drivers gave off quite a bit of heat, due to cooling the TO-3 packaged driver transistors.

Incidentally, the TO-3 is a diamond-shaped package (when viewed from the top), about 1.5 X .75 inches in size.

In the 1980s it was probably the highest heat-dissipation package available. Due to the difficulty of dissipating heat in avionics, the TO-3s were conduction cooled. This meant that the circuit board had a beryllium-copper heat sink laminated onto it and the ICs as well as the driver transistors were mounted through it. The heat sink was thermally connected to the ICs with heat sink grease and to the chassis with "Wedge-Lok" ™ fasteners.

This laminated construction was expensive to design and build (plus there were occasional problems with circuit traces shorting to the heat sink under vibration), so the engineering team opted for a different sort of design which, in retrospect, seems more like a "Rube Goldberg" special than a legitimate design.

The TO-3s were bolted and soldered to the board through a small heat sink block that just surrounded the TO-3 package and was slightly taller. A heat sink plate was then screwed to the heat sink block which was itself fastened with "Wedge-Loks" ™ to the chassis. A "Sil-pad" ™ was used in each of the junctions to promote heat transfer.

This design was marginal due to the thermal conductivity of the extra interface between the heat sink block and the heat sink plate. To add insult to injury, the heat sinks were joined with stainless steel screws mated to Helicoil ™ inserts in the heat sink block. Although the thermal design simulation looked OK, I rather doubt that anyone seriously looked at the screw/ Helicoil ™ contribution to thermal resistance.

All of this analysis is in retrospect, however. The actual failure was kind of scary. While the aircraft was tested on the ground with piped-in chilled air, everything worked OK. The problem occurred when the aircraft was at altitude. Commonly, aircraft cabins are pressurized to an 8,000 foot altitude (this rarified air makes cooling more difficult) and this is what ultimately caused the reported failure. The equipment operator reported smoke in the cabin, the equipment was shut down, an in-flight emergency was declared, and the crew went on oxygen while they tried to vent the smoke from the cabin.

The "slagged" mass of circuitry was evaluated and the following conclusion was drawn: the TO-3 package had heated up sufficiently to melt the solder used to connect its leads to the circuit board. The solder then ran down and shorted the high-power driver transistors, causing burning of the circuit board laminate. The problem was alleviated (as opposed to "fixed") by soldering the TO-3 leads with high-temperature solder (90/10 Sn/Pb). Additionally, the Sil-pads ™ were upgraded to ones with lower thermal resistance, and the heat sink screws were tightened routinely. A routine fix for a potentially deadly problem!!!

Dwight Bues is a Georgia Tech Computer Engineer with 27 years experience in Computer Hardware, Software, and Systems and Interface Design.