Earlier this year, John Shalf, department head of Computer Science at Lawrence Berkeley National Labs (LBNL) presented a plenary presentation entitled “The Future of Computing Beyond Moore’s Law” (SEMICON West, July 2019). The premise espoused was one which you have read many times in IFTLE, i.e. now that CMOS scaling has come to an end: What’s next, and where are we going? Exascale computing, apparently.
Exascale computing refers to computing systems capable of at least one exaFLOPS, (a billion, billion) calculations per second. Exascale computing is considered to be equivalent to the processing power of the human brain at the neural level.
The US will reportedly have its first exascale computer, Aurora, on line in 2021. Aurora will be built by Intel, Cray, and Argonne, through a $500 MM DOE contract. It is part of the National Strategic Computing Initiative.
With thread performance, clock frequency and power all leveled off for more than a decade the question becomes “then what?”
LBNL envisions the decade after the introduction of exascale (i.e. the 2020s) as a period where advantage will be gained by “more efficient packaging and architectures” providing further support for the IFTLE premise (see IFTLE 425) that we are entering the decade where packaging is King.
Post CMOS they see things being driven by “extreme heterogeneity” as shown below.
They see the data movement and technology integration challenges being addressed by photonics and advanced packaging. In terms of options for post CMOS new device technology, Shalf offers the following:
We would obviously like to have a solution that was (1) near term, (2) low complexity, (3) low risk, and (4) high opportunity. As we have said before, it is not clear yet which solution that is.
Will Incandescent Bulbs Make a Comeback?
Those of you who follow IFTLE over the past decade know that I have not been happy with the pseudoscience and politics used to phase out (i.e. ban) the incandescent light bulb through the lighting efficiency standard 2007 which was implemented in 2012. Since we last discussed the fraudulent advertising used for LED home lighting (claiming that the bulbs would last 25,000 hours) in IFTLE 414, it seems that almost daily I find further examples of failed city street lighting or consumers asking why their 100W LED bulbs are failing at a fraction of their promised lifetime.
As an example enjoy the following YouTube analysis of a failed Cree LED bulb that failed in 6 months.
You’ll certainly recall I made this case for Lester the Lightbulb (General Electric’s mascot and my symbol of the incandescent light) by saying the longevity of the LED chip was irrelevant because “It’s the circuit stupid!” If you watch the video, you’ll see that the LED chips didn’t fail but that didn’t matter because these bulbs are circuits and all the parts must operate or the bulb will be limited by the weakest link. In this case, it was the 2.4-ohm resistor that failed. In other cases, it has been the driver and in yet others the electrolytic caps. Anyone who doesn’t think that DOE management knew this when politics drove their decision is, pardon the pun, “not the brightest bulb in the box”.
A recent Sept 10th headline in the Wall Street Journal reads “Long live the Incandescent Bulb” [link]. This article goes on to say that the Energy Department now concludes “….while high-efficiency lights like LEDs can reduce energy costs, their up-front costs remain higher [even after the cost of an LED bulb has dropped from $50 to $2 over the past 8 years] Depending on the light fixture, consumers may not make up the purchase price for years if at all. ….consumers [can] do their own cost-benefit analysis including the functional and aesthetic trade-offs. A homeowner in New York where electric costs are among the highest in the country and utilities subsidize efficiency improvements may make a different choice than a renter in Dallas.”
I know I will be doing my part to bring back Lester. In fact, the only LED bulb I have purchased in the last 8 years is the test bulb that I have been reporting on since that time. Yes, I was one of those who hoarded incandescent bulbs. As IFTLE has espoused previously, saving energy is a great thing to do, but replacing home light bulbs is not how to do it. Energy savings, i.e. your monthly electric bill, is controlled by things with motors, not lightbulbs. More savings will come from an energy-efficient air conditioner or heater or refrigerator. Welcome back, Lester!
For all the latest on Advanced Packaging stay linked to IFTLE…………………….