Link to Original Source
Link to Original Source
I dissected customer returns. Again and again, the products in later shipments looked identical on the outside, but were "cost reduced" on the inside. For example, I would see empty places on the circuit boards where the filter capacitors were supposed to go. In one batch of one product, many of the units were dead on arrival, on the ones that worked when I unpacked them, the solder joints only lasted a few weeks. Once opened, I could see that the boards were either soldered at the wrong temperature, it was the wrong type of solder, or badly made solder. Every connection was visibly a cold solder joint. Either the factory had no quality control, or they ignored the quality control.
Other products looked identical inside and out, but based on the failure rate, the factory must have gotten a bad batch of one the components.
Even longer ago, I worked on a product that logged data to a Compact Flash memory card. It was an embedded product that needed to work across a wide temperature range, including in the winter in Minnesota. The big names like SanDisk would randomly swap component suppliers. Our largest customer saw less than a 2% failure rate, but that was way too much. We found a specialty supplier that charged 5 times as much, but they had a rigorous quality control process. They paid attention to the specifications. They tracked where parts came from, and promised that we would be able to test sample units if they needed to switch suppliers. Alas, the 2% failure rate from the earlier parts had already doomed that product line.
These will be used in data centers where it is common to have redundant systems connected with redundant cables, in order to maintain really high uptimes. Say a hypothetical system has a cluster which consists of 16 compute nodes and 2 storage nodes, Each of CPUserver01 through CPUserver16 will have two of these cables going to storageServerA, and two going to StorageServerB. For a total of 64 of these cables, for that one little compute cluster. Which would leave it an island, so of course there will be more network interfaces.
For this technology to get any market penetration, it will need to be cost effective at these bandwidths, and fit in the racks. Historically, Dense Wavelength Division Multiplexing, DWDM has been great at getting a lot of bandwidth on to a very long single strand (comparatively) inexpensive fiber, which allows in fiber signal amplification, and is the winner at going the distance, but not so good at being cost effective, or space efficient. These things, with the associated drivers should take up far less space inside the servers, and cost less, but they only will get 800Gbits in each direction, only go 300 meters, and use much more expensive (per kilometer of cable) 64 strand fiber.
Rigid silicon requires rigid interconnects. Flexible ICs allow flexible packaging, or different packaging. Instead of building from the printed circuit board up, build from the heatsink up. Use a precision pick and place system to glue the thin, wimpy, inexpensive silicon to the strong massive heatsink. Then mask on the solder balls. Then apply a thin, wimpy, inexpensive circuit "board". Attach all the old style surface mount components to the other side of the circuit "board". "Board" is in quotes because it would get all of its mechanical strength from heatsink. It might be so thin, it is no longer board like.
The big win here, is that one wafer is good for at least 5 sets of circuits. The lose is the grid of holes etched through the silicon as part of the pealing process. Assuming the grid of holes doesn't use up a significant portion of the surface area, the factory is getting close to 5 times as many devices out of each ingot of silicon.
This wouldn't be so embarrassing if the weather service would just delete all that old incriminating information.
Most inventory systems have a way to track cost of goods, age of inventory, and expected profit margin. Eventually retailers will start filling in those details, and tracking them, so they can notice when something goes expensively wrong.
The useful features are
- no connector to wear out,
- alignment is simple.
- The USB/thinport connector is available for other uses. (More of a theoretical benefit, as I don't use the USB port for anything, but I could if I wanted to. I've got the cable, I could even plug in an SD card reader.)
It got worse recently, making it easy to diagnose. It got to the point where the video went away within a second of starting the treadmill. It is an EMI issue. Either the treadmill is emitting too much, or the ComCast box's suicide circuit is too sensitive.
I am so pleased that my ComCast has a suicide circuit to protect me from evildoers who modify treadmills to steal valuable copy righted material.