Ultimate Screw-ups
Comment
Patk0317
Then there is the flip side of the coin. Using devices that are nearing end of ...
WKetel
When the small company that I worked at hired another EE to help with the ...
A case for not choosing the latest components
8/17/2010 9:31 AM EDT
An engineer's tale shows how quickly the leading edge can become the bleeding edge when new components are unexpectedly cancelled or never produced
By Dwight Bues, EE Times Guru
The 1980s were the halcyon days of circuit board design. In the days before Gate Arrays were inexpensively available with sufficient gate counts to completely implement the desired functionality, many manufacturers were producing ICs that implemented functional blocks that could be included in widely varied designs.
Since I was employed designing VME Bus circuit boards at that time, I recall having great flexibility in implementing my designs.Enter the engineer whom we called “Mr. VVI” behind his back.Incidentally, VVI stands for Vanished Vendor Item.A VVI is a device that, although you verified its availability with the vendor, they either never actually did produce or unexpectedly cancelled.
Mr. VVI seemed to have a knack for choosing components that would end up being cancelled.While he teased everyone ELSE for being a “dinosaur,” by using tried and true, or in his mind unimaginative, elements in their designs, he cost project after project tens of thousands of dollars in redesigns for removing the vanished vendor items.
At that time, we actually had a special maintenance contract to provide VVI replacements on a few systems that had been deployed for over 10 years.Here’s something odd that I saw: I may not be the most imaginative designer, but I can’t see why anyone would base a sequencer design on Johnson (ring) counters or J-K flip flops! Although the designer may have known what he was doing in his design, it is unlikely that anyone who came after him would have a clue.
On this project, I employed many Programmable Logic Devices (PLDs) and especially utilized those device sockets you can use to swap the pins around (in those days, it was quite common for different vendors to build similar ICs with different pinouts).Later on, the vendors standardized pinouts of popular items such as PLDs and FIFOs, but you still had to read the data sheets carefully to see if there were any differences.My favorite VVI fix was when I implemented a baud rate generator in a single 24-pin PLD.
Back to Mr. VVI: while we are on the subject of FIFOs, there was a lot of competition in the late 1980s among AMD, IDT, Cypress, and others for who had the fastest FIFO.Mr. VVI just had to use the fastest one possible, which, as I recall, was 25 nanoseconds read time (the rest of us got by with 50 ns. Since he had received sample devices, he decided to put sockets on his board (he would need them!!!).
With his circuit card populated he proceeded to testing.Although all of the control signals seemed correct, he was not able to get any output from the FIFO. After a couple of weeks, he took the chip over to the reliability lab.They popped the cover off the chip and discovered that there was no die inside! In the two months from when he received the samples to when he discovered the empty chips, the manufacturer was able to get the chips to work, and was subsequently able to deliver him functional units, so Mr. VVI got a “mulligan.”
Mr. VVI also utilized a “PLS” or Programmable Logic Sequencer in his board design (I don’t recall the vendor).After many false starts with the logic design software and issues with getting the chips to program correctly, he had a functioning design.Most of the engineers in our group were rather impressed, as it was far faster than most of our other designs.He wasn’t able to rest on his laurels for too long, though.Apparently, the PLS was either too expensive or problematic for the manufacturer to produce.When we went out for quotes on the production quantity of these devices, the manufacturer announced that they were no longer available.A lengthy redesign cycle ensued, since the functionality that the PLS provided was not available elsewhere.
Year later my former boss paid me a compliment.Since our contract turn-on had been delayed, I had left the company by the time my design went into production. He was the one that had been tasked with getting my design up and running.He said that my designs were easy to troubleshoot.That is feedback that I will always cherish.
Dwight Bues is a Georgia Tech Computer Engineer with almost three decades experience in Computer Hardware, Software, and Systems and Interface Design.
|
|
|
|
|
jimcondon
8/17/2010 10:34 AM EDT
Dwight makes an excellent point. In our marketplace, we are often dealing with chips in design or just sampling. This has caused us grief on multiple occasions when the chip hasn't come out on time or even at all. While sometimes it's not possible to avoid this, whenever possible getting chips in production and with second sources at least minimizes the issues and risks to the design.
Sign in to Reply
Dwight Bues
8/17/2010 11:13 PM EDT
Interestingly enough, for much of my career, we didn't have design cycles long enough to build prototypes. Although many designs could trace their ancestry back to previous projects, none were "cut and paste", as it were.
Add to that the risk of VVI components and you REALLY wanted to make every design "brain dead" simple....
Sign in to Reply
kdboyce
8/17/2010 4:15 PM EDT
On a past engineering design project I was criticized for using older parts, or shall I say parts that have been in production for several years by more than one source. While perhaps not as elegant if I had used the latest and greatest, the design nevertheless worked just fine, and everyone who had to deal with it understood how the design was put together and could also trouble shoot it and fix it if a failure occurred. That is not always the case using FPGA's, PLD's, and the like - even if the parts are available - unless the design documentation is flawless.
However, I do understand OEM's who do want a product currently in design because it may contain a feature which gives their products an edge over competition. Both the OEM and the supplier must completely be open in communication and understand the risks involved and have mitigation plans in place. This kind of situation should not be looked at as only a customer-supplier relationship, but as a partnership where both parties have a major vested interest in the project's success.
Sign in to Reply
Dwight Bues
8/17/2010 11:39 PM EDT
I once was tasked to build a VME System Controller function on a card. The day before the Preliminary Design Review, I realized that I had not implemented the "System Reset" and "AC Fail" (power-up reset) circuitry. I whacked together a circuit implemented with two one-shots and took a beating at the Design Review, as these components were a key part of the system and needed to be reliable (also their functionality at power-up was likely to be flawed).
I found a discrete transistor design that my Boss had done on a board that we had used for a many years, implemented two of them for the two different time constants, and happily went to the Final Design Review. My Boss flew into a rage at the discrete transistor design wanting to know "...what IDIOT designed that???". I had a lot of fun telling him who!!!
Sign in to Reply
Frank Eory
8/18/2010 12:48 PM EDT
After all these years, I'm still amazed at how many system problems can be traced back to resets. We really make an effort to emphasize this to our young designers -- reset circuits might be very boring and unlikely to jazz up your resume, but they are one of the most likely reasons for your big multi-core SoC to fail.
In the modern era of RTL signoff, it might seem anachronistic to insist on any amount of gate-level SDF annotated simulation, but we still do a certain amount of it -- at least enough to see the thing come out of reset and start behaving normally.
Sign in to Reply
Haldor
8/20/2010 2:24 PM EDT
We did an ASIC in that contained an 8051 core. I designed our first product containing this ASIC and discovered that the power on reset function (from the IP vendor) was unreliable. Luckily our system had a second processor and I was able to reset the ASIC from the second processor.
Sign in to Reply
ttt3
8/17/2010 6:14 PM EDT
I was involved in a project where we had an entire circuit card design completed with a 1000+ pin ASIC (each board actually had two of these parts). The ASIC manufacturer had produced sample chips but nothing in volume. Just before we went to build prototypes, the ASIC vendor decided to discontinue the chip before it hit production. Many man hours of wasted effort in board layout, schematic capture, supporting component selection and procurement, etc. However, we ended up starting over with a different ASIC and the project was successful. Never did know why that vendor did samples but killed the part before it hit production.
In general, before I select a particular part for a board design, I want to see actual physical stock of the part at a well known distributor. If there's no stock, the part has almost no chance of making it in my design. I have also been bitten by "vaporware" where a vendor will produce a datasheet but no actual parts. Pericom is one vendor I've seen do this. While this is hard to avoid for large ASICs and CPUs, it's a good rule of thumb for smaller ICs, oscillators, crystals, etc.
Sign in to Reply
dutchmam
8/20/2010 2:32 PM EDT
This is where a seasoned Component Engineer can save the company a heap of trouble. I have seen this so many times, when some hot-shot design engineer who decides to use a part that has a "preliminary data-sheet" associated with it. You warn them, they get huffy and insist in using that part anyway. Usually (sure, not always) this is followed by a PCN (product chance notice) or discontinuation notice.
This is one of the things a good Component Engineer does: Warn you of potential disaster in your choice of parts.
Sign in to Reply
W1PK
8/20/2010 3:09 PM EDT
It happens. This is one reason I'm very reluctant to design in a component with a production history of less than 5 years. If it's in stock from multiple sources at one of the small-quantity distributors like Digi-Key or Mouser, that gives me a warm feeling too. There are a few manufacturers you really have to watch out for, who will keep a part in production just long enough for you to get through qualifying it, checking out the prototypes, and delivering a pilot run, and then cancelling it when you try to place high-volume production orders. For some markets, such as military aircraft and nuclear power plants, I woudn't dare even consider a sole-source ASIC; with production lives of 10 to 15 years, service lives of 25 to 50 years, and statutory prohibitions against design changes without extensive and expensive requalification, discrete transistor designs are sometimes the only safe option. All hail the 2N2369 and its SMT cousins.
Sign in to Reply
Jerrysc
8/20/2010 4:24 PM EDT
Some things never change! It is SOP for Marketing to issue product releases for instruments, components, or software that does not exist to one degree or another (but is in the realm of possibility) in order to see if there is enough interest to justify persuing the development or production of same. And if the product is ever successfully developed it wont be in production for long if it is not an economic success.
Sign in to Reply
BarryMoss
8/20/2010 5:12 PM EDT
I've seen the opposite problem, particularly on products being produced over many years. The key parts become end-of-lifed forcing an early end to a product's life cycle. It might be fine to restrict passives and simple parts to parts with long production histories, this is going to work with the more complex ICs, which may not have a production life time of more than 5-7 years. Then there is the cost/size issue. I could produce my design using 5 year old FPGAs, but my competion is going to be using the latest designs in production and have a product 4 times smaller which requires 25% of the power and has a major component BOM 25% of mine.
Sign in to Reply
hm
8/23/2010 12:09 PM EDT
I always like to read latest issue of Electronics Design, EDN, EEtimes and other magazines and hunt for latest parts available. I do save and maintain record of them for our future design requirements. It takes lots of effort to find and understand them from your design/application requirements. Also you generally do this in your own spare time as you have other assigned design work in office.
It is little difficult to get initial samples and technical clarifications from the manufacturer for some specifications and or feasibility of odd topology you want to implement. More often than not, application engineers with manufacturer do not understand you or he/she has difficulty in answering. It takes an extra effort to reach to actual IC designer who has all the answers and suggestions. It is tough mental decision to make final selection of part, predict parts future and informing all concerned party about it advantages and risk. Generally it is encouraged from management with due mitigation alternatives.
However, employing this new generation parts do make product state-of-art with all its associated advantages. Many times it is calculated risk, but this is how one can make small organization effectively competes with big established players and give more innovative solutions. I tried to look back on all previous design work over two decades and I find employing latest parts very rewarding. I take utmost care selecting parts when designing military hardware as it has general requirement of product life span of fifteen years or more.
Sign in to Reply
Mi302
8/24/2010 4:21 PM EDT
I have never encountered this situation. But a better, more relevant war story subject would be vendors lying on data sheets or ommiting particular limitations of their products in their presentations and preliminary data sheets. I can think of a couple situations regarding power devices and heat sinking and also a particularly annoying early flash memory part that had an errata sheet almost as long as the data sheet itself!
No, products I encountered (many cutting edge parts) didn't (eventually) not show up, they just never worked as promised because they were obviously shoved out the door before they were properly tested for customer applications. Work-arounds were often very creative, the aforementioned flash part required significant firmware acrobatics and lowered our products performance (but I'm sure our sales guys never talked about it, so it flows).
Sign in to Reply
Jack.L
8/25/2010 11:34 AM EDT
My favourite was an engineer working on an $80,000 product. Needless to say the volumes were not large, measured in the high hundreds to low thousands. The company had been using a processor vendor for several years that was reliable, predictable, with good tools, and delivery (who will go unnamed) that many engineers in the company were familiar with. The engineer, because he liked some of the features of a new processor, decided to change a processor on a control board. That resulted in delays in the product development not to mention frequent purchasing delays. The cost of the new processor $1.23. The cost of an equivalent from the established vendor $1.45. An $80,000 product delayed to save $0.22 not to mention the added engineering and purchasing cost.
Sign in to Reply
Jack.L
8/25/2010 11:35 AM EDT
My favourite was an engineer working on an $80,000 product. Needless to say the volumes were not large. The company had been using a processor vendor for several years that was reliable, predictable, with good tools, and delivery (who will go unnamed) that many engineers in the company were familiar with. The engineer, because he liked some of the features of a new processor, decided to change a processor on a control board. That resulted in delays in the product development not to mention frequent purchasing delays. The cost of the new processor $1.23. The cost of an equivalent from the established vendor $1.45. Delayed millions in sales to save $0.22.
Sign in to Reply
NevadaDave
8/26/2010 12:07 PM EDT
Our products are not really "cutting edge" - we don't need the fastest or smallest IC's. What we need are good, long-term supplies of the parts we DO use, and that has been problematic. We recently had two different parts go obsolete, and there are no equivalent replacements available for them. Since we don't do high volumes, the per-unit cost of re-designing using available parts is high, so we are in the dilemma of trying to decide if it's worth keeping the product around or not.
Sign in to Reply
WKetel
9/3/2010 5:53 PM EDT
When the small company that I worked at hired another EE to help with the workload, it became interesting. Fairly soon he pointed out that he could greatly simplify the circuit board of one of our highest production quantity products. This could shorten assembly time and make the circuit board smaller. I asked about the cost, and he explained that for a circuit that simple the engineering would be less than $50,000, so it would be a real bargain. Then I pointed out that the total price of all 5 ICs was about $3.50 a unit, and that a very large production run would be 50 units, so it might take a while to amortize the cost of the chip. The other advantage is that all of the chips in the system were available from at least 5 major manufacturers, and at 20 different distributors. So sometimes the custom chip is just a poor choice.
Sign in to Reply
Patk0317
9/8/2010 3:49 PM EDT
Then there is the flip side of the coin. Using devices that are nearing end of life cycle. Not available is not available whether they are the newest or oldest chips on the board.
One of my customers in the 90's was using FPGAs and CPLDs to mimic other disappearing devices. He made a good living at it (for all I know he still does).
Sign in to Reply