GlobalFoundries joins Google’s open source silicon initiative(opensource.googleblog.com) |
GlobalFoundries joins Google’s open source silicon initiative(opensource.googleblog.com) |
Someone's gotta do it
Most of the stock market data the avg person sees is 15 minutes delayed by request of the NYSE, and in a world where trades are reaching the 3 millisecond and below mark, that's a very long time. So Google opening up what could be <20ms data might be able to help some applications. Not saying it's a silver bullet either.
I wish they would take on even more big projects that need to be tackled. Like I want to see a google branded remote controlled barge out in the pacific garbage patch that the public is controlling through a google web app or some other insane, good projects. Idk when you deal with finance companies all day, even google seems moral.
However, it's not looking good this quarter for all tech cos
Whilst I like the thought behind idea, these ocean garbage patches aren't easy to clean up with a barge.
> For many people, the idea of a “garbage patch” conjures up images of an island of trash floating on the ocean. In reality, these patches are almost entirely made up of tiny bits of plastic, called microplastics. Microplastics can’t always be seen by the naked eye. Even satellite imagery doesn’t show a giant patch of garbage. The microplastics of the Great Pacific Garbage Patch can simply make the water look like a cloudy soup. This soup is intermixed with larger items, such as fishing gear and shoes.
Better to clean things up at the source. It is estimated that 90% of all plastic waste in our oceans come from just 10 rivers in Asia.
https://education.nationalgeographic.org/resource/great-paci...
https://m.dw.com/en/almost-all-plastic-in-the-ocean-comes-fr...
And yes, they also started building river interceptors as that is more bang per buck. They still seem to think they can clean the existing stuff economically too.
Cursory search turned up nothing for me. Can you point me in the direction where I can find free (or low cost) market data that you mentioned?
This link has more info
Admittedly, the bones they'll throw us in the process do look like they'll be cool, tho.
Competition is good.
Nit pick: you've got your units mixed up; minute is a unit of time, and light-year is a unit of distance.
Industry average for those colocated and trading using software is going to be closer to something like 20-30 microseconds
You notice that “open” Android is increasingly dependent on closed source “Google Play Services” and Google uses the closed source parts to beat OEMs into submission?
Same with Elon Musk. I remember him being compared to a comic book hero. Now he's an alt-right racist (if you are a leftist) or an evil globalist technocrat (if you are a rightist).
Every company whose business model isn’t “I give them money and they give me stuff” is shady.
He hasn't changed though, and little has changed in what he's actually doing too. If anything, he's more mellow now than he was when he was younger. People's impressions of him have been highly twisted IMO.
as well they should be, they carried it forward like Atlas carrying the globe.. in the beginning it was competition, then, we could not stop..
With this announcement and the other SkyWater 90nm they're bringing out (https://opensource.googleblog.com/2022/07/SkyWater-and-Googl...) it looks like they're keen to fund this for the long haul which is fantastic news.
I think there's great potential in open silicon but precious little of it around right now. Open tooling is still in early development stages in particular for implementation flows (i.e. actually producing a silicon layout from your RTL design). Development here is stifled by how closed this world is. Open PDKs, that can be used for real chips, are a great boon and I hope will really accelerate developments.
efabless.com makes custom short run chips, right? Does this mean Google covered the cost of 240 chips for other people?
$2.34m.
That's kinda awesome.
Last week SkyWater announced a 90nm PDK, which is smaller than the current SkyWater 130nm. Also crucially, this one is FDSOI (fully-depleted silicon on insulator), which keeps chips from leaking power, meaning this should be much more ideal for any kind of mobile always-on design. Neat.
And now comes GF's 180MCU PDK. At first, it's like: this is bigger, what the heck! But it's designed for mixed signal & power designs, like Power Management Integrated Circuits (PMICs).
Really neat having something a little smaller and something a little bigger right near where we started. I really hope some of the existing people who've done open chip design on the existing shuttles get to make similar-ish chips on the new bigger & smaller processes & can nicely characterize some of the performance differences.
This all just makes so much sense. Making sure chipmaking is a healthy, robust, creative, growing industry is in everyone's interest. Without efforts like this, it feels like the industry is at risk of too much consolidation, of everyone getting bought out and/or giving up, and there not being vibrant chipmaking at all. The wave of buyouts across the 201X's was shocking. I continue to think part of the chipmaking crunch was just that there are so many less independent entities now; the diversity of the ecosystem is way down.
The transistors on this process are very slow and power-hungry, even for such an old process -- the lowest rated supply voltage for this process is 3.3 volts whereas 1.8 volts is typical for 180nm. That's ~3.35x the active power consumption and slower switching speeds due to higher threshhold. It's meant for people who had 350nm designs (mid-1990s) and want to migrate them with the minimum possible effort (nearly all fabs have shut down their 350nm lines).
It is totally awesome that Google were able to convince GF to release this PDK! But they should be more up-front about the fact that this is a High Voltage process; most people won't notice the "MCU" at the end of the process name or know what it stands for. Their announcement about a 90nm process with Skywater last week is a big deal; this not so much, unless it's just the first of a series of process releases from GloFo.
Also this does make this 180 nm variant much more useful.
There are many kinds of interfaces or analog devices for which such a high-voltage process is absolutely needed.
Now, you can design a complete chip set for certain applications, e.g. with a central chip designed with the 90 nm Skywater PDK and one or more peripheral chips designed with the 180 nm GlobalFoundries PDK, for power management or various interfaces.
Are you sure about that? The mandatory "Caravel" pad ring occupies the entire top layer of the chip as well as a ring around the edges. It hermetically entombs the design area that you control.
There is no exposed surface under your control for any kind of chemically-relevant device like a microwell.
Correct me if I am wrong, this means I could do this. Perhaps I am wrong! In which case I should find a different chip supplier.
The hope here is that open sourcing the design rules will help build an ecosystem of open source silicon. While it's true that 180nm is too big for anything high performance, it's perfectly fine for hobbyists who don't have the money required for more advanced nodes anyway. On the foundry side, they don't need to be as secretive about old nodes and it might increase demand for underutilized manufacturing capacity.
180nm isn't crazy awful though, is it? That's about 20 years back, so you won't be doing ML, but it's enough to have industry applications and low power general purpose computing.
at this stage, presumably no one would risk the capital and time required to copy the foundry while publishing details increases the likelihood of attracting more clients.
thanks in advance for the clarification on why the industry operates the way it does.
china is adamant about taiwan, like putin about the donbass. So it is not unreasonable to think that what is happening for the donbass, will happen for taiwan sooner or later.
maybe it is time for top-notch foundries to move out (and the ppl who would require more freedom that the china regime tolerate, like in hong-kong).
Distance/time * time = distance
Waiting a "light year" would be a year.
Yes, you are correct that astronomers talk about seeing into the past by X number of years when we look at a galaxy X light years away.
While you would not want to make a CPU in 180 nm now, or any purely digital circuit, which could be better made with a FPGA, unless you need clock frequencies over 1 GHz, if you want to make any mixed digital-analog circuit with an important analog part, 180 nm can be fine.
Any analog circuit part made in 180 nm will not be much larger than in any up-to-date process, because the dimensions of the analog components are determined by functional requirements, such as noise or maximum current, not by the lithography limits.
Also, to the point about Bloomberg's high subscription fees, they provide all imaginable financial data under the sun. This google demo with CME is just CME data, which is a small, albeit important, part of market data, let alone the broader financial data. Furthermore, in many cases the data itself is owned by the exchanges, so Bloomberg's fees reflect passing on the (at times exorbitant) exchange fees. NYSE, NASDAQ, CME, etc.. are historically the ones that are the gatekeepers of the data and putting up financial barriers.
And yes, you're right on these costs. Some exchanges will charge large 5 figure sums annually just to give you a port so you can subscribe to some unreliable (UDP), poorly thought out, over-engineered-in-the-early-2000s, market data protocol that requires you to spend weeks of engineering effort to debug and normalize because hey... every exchange wants to use a different tech stack and none of their customers want to use a vendor library, so they don't exist.
The irony is a lot of organisations that are used to dealing with these feeds struggle when it comes to crypto, because they're not used to a world of websockets and public facing IPs, or the risk of exposing their machines to the internet.
One company I know pays a small 5 figure annual sum to a shitty "cloud solutions provider" for what is essentially just a dual NIC reverse proxy providing access to Coinbase via a stable IP. The is provider of course charges thousands per IP. Managing public cloud infrastructure is just outside of its comfort zone.
"This innovative collaboration with Google Cloud will not only make it easier for our clients to access the data [...] CME Group customers will be able to access all real-time CME Group data [...]" (emphasis mine) [0]
As for [1], it has a Contact Us form and a link to [2] which talks about onboarding and an "On-demand, pay as you go model".
Could you point to the free market data pub/sub and/or specific steps to recreate the demo? Thanks!
[0] https://www.prnewswire.com/news-releases/cme-group-to-offer-...
[1] https://www.cmegroup.com/market-data/connect-data/cloud-mdp....
It would be interesting to see $ spent per tonne of garbage removed from the rivers vs ocean. Like you said, far more economical to focus on the source. Given no project has unlimited funds, why spend resources on the"garbage patch"? I suspect that they still work on the ocean part because:
1. Is better for fundraising;
2. Is a fun engineering challenge;
3. fun to work on big boats;
4. Helps with recruitment;
Check their Interceptors.
https://theoceancleanup.com/rivers/
Also, the ocean garbage does need cleaning up too
True. But I'm also guessing that in 5 years I'll be more frustrated at the level of control they're exercising with thing they're currently building next to it. More so than I am with the market they're commoditising today.
And you get the worst of all deals, see Shrinkflation
My point is we shouldn't trust businesses to be moral, and I don't. I'm just angry at the NYSE
I'm not suggesting Google gives away information as a purely altruistic endeavor. However, Google's ulterior motives, should they have any, are irrelevant to the value of the information itself so long as it hasn't been edited or tampered with.
It's also worth mentioning that lots of businesses that you pay money to, track you in similar ways. The origin of it was probably supermarket loyalty cards in the pre-internet era.
The association with no-cost products is not really appropriate, and mostly seems to be progaganda from people with other business models. It was the web scale and pure digital products that let the cost drop to near zero, which synergises with ad-supported offerings, (just as "free" ad-supported TV and radio were a thing for decades). However, it also allows other business models, such as non-profits like Wikipedia, netflix subscriptions, or subsidized by necessary hardware purchases.
Similarly, just because you pay money for various Google services, doesn't really gave them any motivation to not track you.
The only time that ads don't really bother me are the rare occasions I watch sports. The breaks are more natural.
The padframe and top metal layer which "entomb" Caravel designs are both part of the die. They are not part of the package. Omitting the package does not help you here.
You might be confusing the padframe with the leadframe. The padframe is part of the die, the leadframe is part of the package. In a QFN package they are connected to each other by extremely thin gold wires.
"From the ChipIgnite program you can get bare die; from the Open MPW program, you can't. If you submit something on a ChipIgnite shuttle run, then the pads will be exposed, along with any other places on the chip that you place overglass cuts (i.e., you can place your own pad cuts internal to the user area)."
The other problem is that even for $10,000 the ChipIgnite program won't give you a confirmed booking. You have to submit your design and then they decide if they feel like manufacturing it. Any chip worth paying $10k to fab is going to involve an engineering sunk cost of around 5 times that amount. Which is why Real Fabs take your money and give you a confirmed booking.
Even if it's not bleeding edge, there still might be special features or secrets they want to protect. For really old nodes, I think it boils down to two reasons: 1) Clients with money don't care about open design rules, so there was basically no demand for it. 2) Foundries generally have a pretty strong culture of secrecy, so without someone asking externally they're not going to open source anything.
assuming no one will invest the time and capital to build a similar foundry, is it fair to say that such secrecy has minimal benefits? or can an existing foundry learn these secrets and improve their own products pretty easily?
I guess let me put it this way - I think it's entirely rational for them to be cautious. There's currently not a lot of upside to being more open, so it's usually not worth it even if the risk is small. Programs like this are great because they hit both sides of the equation: increasing demand for open PDKs while lowing the perceived risk by building a track record of successful releases.