Multi-die system or chiplet-based technology is a big bet on high-performance chip design:
While the global semiconductor shortage that began in 2020 had its proximate causes in natural disasters and geopolitics, its effects drew widespread attention to the fact that just about every industry relies on chips. And pandemic-related ripple effects aside, the silicon status quo has been in flux for some time. New technologies like artificial intelligence and machine learning (AI/ML), which require greater computing efficiency and performance, have strained traditional systems in recent years.
With the rise of the Internet of Things (IoT), customers have also come to expect intelligence in everything from refrigerators to lightbulbs. Innovators are responding accordingly. Our poll found that nearly one-third (31%) of business executives plan to improve upon their companies' existing smart products, and almost another third (29%) intend to add AI/ML capabilities to their products soon. Only 9% of respondents said they were not producing IoT or connected devices.
This type of technology, however, necessitates robust edge computing and on-device processing, which requires greater and more efficient hardware performance. Complicating matters, the cloud data centers powering this compute shift are also voracious energy consumers. This is another area where traditional silicon is stagnating: sustainability. The cost of producing superfluous silicon is not just bad for business—it has an environmental impact. And while there's an ongoing push toward net-zero carbon emissions within the semiconductor supply chain, the industry isn't yet on track to meet the emissions standards set forth in the UN 2016 Paris Agreement.
An industry shift toward multi-die design could be part of the solution to these challenges. Instead of a single monolithic chip ("system on chip"), multi-die designs consist of a collection of chips (chiplets or dies) linked in a sophisticated package ("systems of chips"), which can include stacking blocks in a 3D configuration for greater density. Multi-die system designs are capable of supporting the rollout of AI/ML at scale, and they can improve silicon yields, reducing waste during chip manufacturing.
When it comes to the business use cases for multi-die systems, Patrick Moorhead, founder, CEO, and chief analyst at global technology consulting firm Moor Insights & Strategy, notes that these custom designs may soon be a key differentiator for companies looking to stand out among competitors. "As more people are looking at more custom silicon as a way to differentiate what they bring to the table, that's what businesspeople should be looking at," he says. "Chiplets enable smaller companies with smaller pocketbooks to use semiconductors for unique competitive advantage."
(Score: 4, Interesting) by krishnoid on Monday April 03, @12:22AM (3 children)
I can't be sure, but I think that Intel dude left Intel because he couldn't do what he wanted to do there, then he helped put together Apple's M1 and M2 multi-die-in-package processors. Then he went back to Intel [macrumors.com], and I'm guessing it's because they lured him back with promises of letting him do the same thing at Intel. After Apple and Intel, you gotta figure the rest of the industry will try to play catch-up.
(Score: 3, Insightful) by Anonymous Coward on Monday April 03, @02:55AM (2 children)
You make it seem as if Intel/Apple is the leader in such stuff. AMD was doing it before Intel: https://www.pcgamer.com/intel-slide-criticizes-amd-for-using-glued-together-dies-in-epyc-processors/ [pcgamer.com]
(Score: 2) by krishnoid on Monday April 03, @04:04AM
More along the lines of brand-name recognition and marketing.
(Score: 3, Interesting) by takyon on Monday April 03, @09:35AM
TSMC: the real leader
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 4, Interesting) by JoeMerchant on Monday April 03, @12:49AM
Worked with a hearing aid company (Valtronic) some 25 years ago, they were all into multiple die assemblies back then, for the extreme miniaturization aspects. It was a bit pricey at the time, but they had customers willing to pay...
Україна досі не є частиною Росії Слава Україні🌻 https://news.stanford.edu/2023/02/17/will-russia-ukraine-war-end
(Score: 3, Funny) by hendrikboom on Monday April 03, @02:13AM (2 children)
Why would I want AI in my refrigerator? I don't even want the damn ice machine taking up space in its freezer. A few ice cube trays filled with water (which turns to ice) will do as well, without a lot of space-consuming machinery. There's not enough space left for food! I have trouble fitting a frozen pizza into the freezer. Not to mention that the port in the refrigerator's front ice-cube port keeps getting blocked by frozen condensation.
(Score: 4, Funny) by maxwell demon on Monday April 03, @03:39AM
You don't want your AI to be hot-headed, so you better put it in the fridge.
The Tao of math: The numbers you can count are not the real numbers.
(Score: 2) by DannyB on Monday April 03, @06:12PM
Which companies are that 9%? Those are the companies I want to buy non IoT products from!
How often should I have my memory checked? I used to know but...
(Score: 5, Insightful) by Rosco P. Coltrane on Monday April 03, @05:01AM (4 children)
Not quite: people expect to find "smart" in almost every electrical appliance nowadays as something unavoidable, just like they expect to fart when they eat refried beans: they don't want it, they just know it's gonna happen and there's nothing they can do about it.
As for intelligence, I don't think anybody expects any in IoT devices. Let's net get ahead of ourselves here.
(Score: 3, Interesting) by JoeMerchant on Monday April 03, @10:27AM (3 children)
How intelligent should a refrigerator be?
At least a thermocouple in one compartment as a thermostat and a (possibly manual) airflow control between the compartments for temperature balance. Oh, and a pin switch to turn off the light when the door is shut. That has been state of the art refrigerator intelligence since the 1940s through to the 2010s. We bought a new fridge in 2013 and that's how smart it is.
We had a smarter one from 2004, and it's "main board" died in 2007 effectively "totalling" the fridge: $600 repair bill plus a week of no refrigerator waiting for the part. Instead we replaced it for $850, ordered the board off the internet for $160, waited three weeks for the board to arrive, self-replaced it then sold the repaired fridge used for $425.
Now, I would appreciate thermosensors in both compartments, a digital (or even analog dial) readout of temperature in both compartments, set point by actual calibrated temperature instead of a 1-7 scale knob, and maybe a "door is not closed" alarm after a couple of minutes. With LED lighting the switch off when closed is almost irrelevant, but you will have pin switches for the door open sensors anyway, so why not? Anything else? Put it in a standard Android tablet that magnetically mounts to the door. Wireless connectivity? Not to the fridge itself, thanks, just WiFi for the tablet.
There is absolutely no reason to introduce a cyber vulnerability that might cause my food to spoil, none.
I will say: the camera / screen that shows fridge contents when the door is closed is pretty slick, maybe having that available on WiFi would be worthwhile (check fridge contents when you are at the store?) and better than a simple glass panel, but that system should be 100% incapable of affecting cooling functions.
I will also say that the current double door craze is functionally inferior to a simple gravity powered self closing single door, and anyone who suggests a power servo door closing mechanism deserves a long Siberian vacation.
Україна досі не є частиною Росії Слава Україні🌻 https://news.stanford.edu/2023/02/17/will-russia-ukraine-war-end
(Score: 3, Insightful) by khallow on Monday April 03, @11:33AM (2 children)
(Score: 4, Insightful) by JoeMerchant on Monday April 03, @04:14PM (1 child)
>No customer need is served.
While I agree, follow the pretzel logic of the weekly Marketing meeting:
"Most customers want a lower up-front cost. Our research indicates 4/5 Big Box customers will buy a $500 refrigerator with a $10 monthly payment before they'll buy a $800 refrigerator, and on average they will allow that $10 per month subscription to run for 6 years before they terminate it. Giving the first year of subscription for free reduces objections to the subscription cost by 60%, and quoting the subscription cost at $3.99 per month but then raising it to $7.99 per month after 2 years and $12.99 per month after 2 more years also nets higher overall returns. All we need to keep the subscription running is one 'killer app' for the decision maker, which could be anything from a music subscription service to a home security camera, or even something vaguely related to the refrigerator itself like a scrollable zoomable view of the inside of the fridge on their cellphone."
Software as a service: the bigger the company, the smaller the amortized cost per unit. And then you have players like Disney+ who can't even be bothered to get their streaming interfaces up to Netflix's low bar standard after years in development and the live market, but they've already made that $7.99 to $12.99 price hike. Disney+ on your fridge, right? Gotta be some moms who would love to play Aladdin and The Rescuers to their newborns from the fridge screen to keep them settled in their high chair...
Україна досі не є частиною Росії Слава Україні🌻 https://news.stanford.edu/2023/02/17/will-russia-ukraine-war-end
(Score: 1) by khallow on Monday April 03, @07:03PM
This is a good point. Low costs up front with backloaded higher costs later. Still a bit surprising that nobody has bothered to find anything really compelling to lure the suckers in.