How does a chip go from being an academic theory to a $10 billion R&D Project?
In: Technology
I imagine that once they actually develop something, all the bills get dumped under the expenses of that project, including all the restaurants, spa’s, conferences, hookers, cars, yachts, golfing, and everything else. It builds up quickly.
Let’s say you want to make clay bunnies with more and more details like hair etc. You’ll need to work on the clay and bake it in the oven. However, you find out, most clay doesn’t work for the fine details like hair and it all melts/becomes a mess. You figure out you need different clay and work really hard on it and try all the settings of your oven. Eventually you even have to build your own oven to get the best clay bunny in world.
All that experimenting with clay and ovens can be considered R&D and it can cost a lot of money because you end up with a lot of ugly/failed bunnies and one perfect bunny. After that, making the perfect bunny becomes easy because you know what clay, how much water and you have the perfect oven allowing for mass production of bunnies at lower cost.
Also, the CEO needed a new house, and some of the heads of R&D probably put down coke as a business expense
R&D isn’t a job title, it is a broad classification. Many jobs can fall under the title of R&D. Scientists, engineers, product designers…etc
Imagine I run sandwich factory. My R&D workers may be experimenting with mixing new ratios of flour and water for new bread types, or they may try new combos of meats, cheeses, and condiments.
Once they come up with something tasty, the work passes to the factory line, and they produce the sandwiches, and other people in the company go out and make advertising and sell the sandwiches.
TL:RD. The R&D people are inventors, and anyone who works on creating a new technology or product. Everyone else in the company produces, maintains, and sells what the R&D team ivented.
Lets say you own a large metal mug. In fact the whole world owns only large metal mugs. But you want a small metal mug with a rubber lid. How do you go about that?
You need to research if it is even theoretically possible to make mugs smaller. And figure out if the rubber is going to do anything to the mug, or the contents of the mug. Maybe you THINK you want a smaller mug, but a medium size mug would work fine. Is there even a market for the mug?
And if everything is pointing to “yes, you can have your mug” you’ll then need to figure out how to create the mug physically. If the whole world has only big mugs, you’ll need to create machines that will create the smaller parts. In order to do that, you’ll have to go back and research those machines….and is that going to cause the machines to fail faster or produce defects in the mugs? How do you fix those issues?
And those machines might have the need to create new machines to create THEM. So you research some more. You can see how this might create a chain of R&D….but then you find out the rubber can only be sourced from one type of tree (lets call it the small mug tree or SMT for short). The SMT can only produce 1lb of rubber a year, but you need 1000lbs. So you might have to buy more land to harvest the SMT. The locals hate you though so you have to jump through a bunch of permits and town meetings etc. You might even have to build a road and a small port to ship the rubber. That costs money and time.
If you want more information about the “connections” of all of these technologies I highly recommend James Burke’s “Connections” series. It takes a MUCH larger view on how something 500 years ago gives us space flight, but it gives you a better picture of what actually goes into creating something “new”. The cost of the chip is cheap, but the money has to flow to get to the point of being able to make it.
The software alone that is used to design/verify chips cost millions of dollars. I used to work for an EDA company and the software used to verify a chip design would cost customers around 250k to $1 million per license. To scale up the software and be able to run on 100s of CPU you would need 10s to 100s of licenses.
Researcher: hey I have an idea for a new chip let’s build a machine that can make it.
Engineer: that’ll be 5 million dollars please.
Researcher: oh darn we reached the atomic limits and are crossing into quantum territory and electrons are clouding across atomic barrier. We need a new machine.
Engineer: 200million dollars please.
And so on until they built enough machines to figure out how do build it right.
They do research math and science, i work at a big tech firm and many of the developers actually have received patents for their inventions, yes there is patents for hardware and software breakthroughs
Just to have an idea, a single machine to make one of those chips costs anywhere between 7 million and 100 million dollars. You need hundreds of those machines, each doing a different step, each working as much as possible, each being maintained by at least one highly paid engineer.
Then you need the people that figure out the machines, that figure out the right materials to put in the chips (me!), that figure out which specific order to perform the steps.
And then you have what most people are talking about, how to arrange the transistors and memory and so on.
*Lots* of expensive moving parts, all working together for the future of AI
R&d encompasses a lot of stuff. Making one off chips is amazingly expensive, so having to test little changes would just rake up a lot of it. You need the engineers to design the chip. Create simulations or models. Even the software development for this a lot falls in this category. They also probably try lots of things that don’t go out, so all there exploratory and failures get rolled in. Hardware to do certain things also just cost stupid amounts of money sometimes. Like one piece of some hardware could cost millions alone, depending on needs.
Also don’t forget people cost money, and I imagine the people at nvidia cost a lot per an engineer like hundreds of thousand each, so if they have like 1000 engineers they probably have 10s of millions in cost there. They also could be fudging the numbers a bit with people that have unexercised stock, since there stock went up a stupid amount, and they definitely made millionaires out of some of there employees, they could consider that some how, if it’s just them saying shit in stage.
Latest Answers