If you are a regular searcher for exotic parts among the virtual pages of semiconductor supplies catalogs, you will have probably noticed that for a given function it is most often the part bearing the Analog Devices logo that is the most interesting. It may have more functionality, perhaps it will be of a higher specification, and it will certainly have a much higher price. [Zeptobars] has decapped and analyzed an AD chip that holds all three of those honors, the AD9361 SDR transceiver.
It’s placed under a slightly inflammatory title, “when microchips are more profitable than drugs“, but does make a good job of answering why a semiconductor device at the very cutting edge of what is possible at the time of release can be so expensive. The AD9361 is an all-in-one SDR transceiver with an astonishing bandwidth, and as such was a particularly special device when it reached the market in 2013. We see some particularly fine examples of on-chip inductors and PLL circuitry that must have consumed a significant design effort to preserve both bandwidth and noise characteristics. This is an item of physical beauty at a microscopic scale as well as one of technical achievement.
The financial analysis puts Analog Devices’s gross profit at about $103 of the $275 retail purchase price of an AD9361. The biggest slice at $105 goes to the distributor, and surprisingly the R&D and manufacturing costs are not as large as you might expect. How accurate these figures are is anybody’s guess, but they are derived from an R&D figure in the published financial report, so there is some credence to be given to them.
We’ve featured [Zeptobar’s] work before more than once. A look at fake Nordic Semi parts for example or a Soviet i8080 clone have received their treatment. Always a source to watch out for!
Analog Devices datasheets and documentation are a thing of educational beauty.
RIP.
Their typesetting could be better, though.
(Disclosure: I’m a FAE in the semiconductor industry) The statements about “Gross Profit” are really misleading- the financial statements contain just the accounting definition of “gross profit” – Take away the Gross Profit and you don’t have a company to produce the part. Yes, the cost of these chips are *very* high, but that reflects upon all the back-end that supports the development, production, and the customers. This chip is expensive because it was very expensive to produce, test, and support.
The semiconductor industry is consolidating right now because of the dwindling margins and lack of (true) profitability. Companies that are truly profitable have such a premium on their price that they are not a target for acquisition.
And then there’s specialization where one does one, or a few things really well, then sell their services to all comers that can afford them.
Hey Matt! I used to do that years ago. We were just at 1 micron and it was VERY hard to actually probe the die in those days. I can’t imagine what you are doing these days. We played with LCD and electron beam stroboscopy but none of that was very practical in those days.
I agree, I think they’re understating the NRE costs for the AD9361 by just using the company’s overall % NRE. I’m pretty sure that this is a low-volume product that has to spread out those costs among not a lot of devices.
So…you want to make it cheaper? Buy more of them :)!
It usually is a balance between the competitors for a given chip functional range class, and actual process cost.
Mergers can temporarily prop-up share values during economic stagnated periods… and I can only surmise AD had an MBA management infestation issue for years…. as the merger disease has taken out a number of companies trying to game production supply chains with cheesy 1800’s styled economics. See the fate of ATmega2560 at its current lot prices from Microchip…
Also, I recall a project early in my career that was reliant on AD specific chip IP… and the unit cost ballooned from $6.50 to $58.00 in half a year. The predicted product production budget was no longer stable… Lesson learned… if you want to keep your job, than you don’t ever want to use vendor specific parts.
https://www.youtube.com/watch?v=3zc4bGkU05o
How much of their “profit” actually includes G&A overhead not specifically attributable to this chip?
Remember when you could get many of AD’s best bits as free engineering samples?
It was definitely interesting to see that the chip is only about ~21mm^2 in a 65nm process. It also looks like it’s a wirebond package (I don’t know enough about RF/analog to really know if flip-chip is that useful for RF). I’ve seen/heard of this chip a lot, but never knew the details about its manufacturing.
Did they figure out how many metal layers were used by Analog Devices?
On the number of metal layers – nope. But I assume they are not really limited by metal layers here, so regular 10-ish metals would do. Inductor X-section would have been interesting though, hopefully someday I will be able to do X-sections.
I don’t think they’re interconnect-limited either, but it would be interesting to see if the chip’s only say 4-6 metal layers.
Pretty
Great photo’s of a cool chip that I’ve done a lot of projects with in the last 7 years. However the cost analysis is basically nonsense, as are the assumptions about ASP and volume, which is not to harsh on the author, who’s die photo’s I enjoy, but more a statement on how hard the semiconductor industry is to understand unless you are a long term insider.
on the subject of Analog Devices, I find myself often needing high speed true 4 quadrant analog multipliers…. it seems that in the last couple of years, the only one readily available in the UK with reasonable performance and availability is the AD633 series (at least that I’ve found). And they cost…. a lot… considering how conceptually (and in terms of exotic analog, in actuality) simple these devices are. I guess they’re not in demand anymore so these prices have to be paid if you’re in a “niche”. Anyone know of any other 4 quadrant analog multipliers still around?
Concerning the drugs, I just imagined a drug cartel wanting to be able to test a batches potency scientifically and needing a chip with a special sensor in it, then somehow producing chips becomes more profitable than drug running
Nah you’d probably use what they use in piss tests. Antibody-based test agents that produce a visible mark, then cheap opto-electronics to decide exactly how dark or thick the mark is. They need re-certifying every now and then.
Unless you’re into secondary standards or a lower volume niche market, NIST and USP/NF dominate the primary standard market.
Gross profit of $103 dollars with the largest chunk going to the distributor is quite significant.