You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
With ever accelerating progress in technology, what is the half-life of excess inventory, when progress is at an ever-accelerating clip? Will there be write downs as excess inventory fall a half to a full generation behind? Any thoughts or comments appreciated.
It's going to be completely different depending on the type of chip.
Commodity DRAM is driven by the memory market and advance of technology; the price decrease of DRAM over time has been well documented. See for example John C McCallum's graph of memory prices: https://jcmit.net/mem2015.htm --- this has actually slowed down since around 2012. (Jim Handy says since 2011.) Before then, it fell by about a factor of 100 per decade, which is about a price cut in half every 18 months. Since 2012 it's **much** slower and the oscillations of the memory cycle are large enough that it's too soon to predict going forward, but it looks more like a price cut in half every decade.
For analog / power: hardly any decrease in value over time. An op-amp or voltage reference or voltage regulator from 20 years ago is just as usable as one today. I don't know how the reliability decreases for chips in storage... 5 years is probably nothing to worry about, 20 years might make me nervous.
For microcontrollers: these are generally long-lived parts with 10-20 year life cycles, so a few years in inventory is no big deal as long as the sales keep going.
FPGAs --- out of my range of familiarity.
For leading-edge stuff (microprocessors, SOCs, etc.) --- that's out of my range of familiarity, but likely it "spoils" a lot quicker.
CPUs and GPUs - I'd actually argue that in historical terms the half life time is growing.
CPU:
In the 1990s / Dennard scaling days, CPUs scaled constantly to the point that one year later a $100 CPU could clobber a $750 one from a year ago. (ex: Celeron 300A vs Pentium II-300 Klamath in ~ 1999).
Today AMD is launching ~ 20-30% improvements every 2 years (+ doubling of cores every 4-5 years) for desktop and mobile. Intel was flattish from 9th through 11th gen, with a ~ 30-40% boost for 12th gen, and another 15% for 13th gen. The last two being sort of impressive but a far cry from the 1990s.
GPU:
With the possible breakdown of the economics of Moore's Law -- the future looks like 1-2 super high end halo products that improve performance by 40-60% (every 2 years), but mid tier / volume products only offering a 5-10% perf/$ improvement over the previous gen. There will be feature improvements that draw people to newer chips, but the half life is longer as Moore's Law slows to a crawl due to economics.
I also think DRAM on average is flat or slowing down -- it's normal to come up with a new DDR standard every 5-6 years now, and prices just fluctuate based on [insert everything]. It looks like memory scaling is nearing the end so half lives should extend time-wise.
IMO, The perception that it's accelerating is due to the varying oscillations in greater market conditions / shortages / perceived shortages of the past few years. Take away the mad-dash to work from home as a one time event and you have the average of.. slowing time.
