NASA Makes RISC-V the Go-to Ecosystem for Future Space Missions (sifive.com) 54
SiFive is the first company to produce a chip implementing the RISC-V ISA.
They've now been selected to provide the core CPU for NASA's next generation High-Performance Spaceflight Computing processor (or HSPC), according to a SiFive announcement: HPSC is expected to be used in virtually every future space mission, from planetary exploration to lunar and Mars surface missions.
HPSC will utilize an 8-core, SiFive® Intelligence X280 RISC-V vector core, as well as four additional SiFive RISC-V cores, to deliver 100x the computational capability of today's space computers. This massive increase in computing performance will help usher in new possibilities for a variety of mission elements such as autonomous rovers, vision processing, space flight, guidance systems, communications, and other applications....
The SiFive X280 is a multi-core capable RISC-V processor with vector extensions and SiFive Intelligence Extensions and is optimized for AI/ML compute at the edge. The X280 is ideal for applications requiring high-throughput, single-thread performance while under significant power constraints. The X280 has demonstrated a 100x increase in compute capabilities compared to today's space computers..
In scientific and space workloads, the X280 provides several orders of magnitude improvement compared to competitive CPU solutions.
A business development executive at SiFive says their X280 core "demonstrates orders of magnitude performance gains over competing processor technology," adding that the company's IP "allows NASA to take advantage of the support, flexibility, and long-term viability of the fast-growing global RISC-V ecosystem.
"We've always said that with SiFive the future has no limits, and we're excited to see the impact of our innovations extend well beyond our planet."
And their announcement stresses that open hardware is a win for everybody: The open and collaborative nature of RISC-V will allow the broad academic and scientific software development community to contribute and develop scientific applications and algorithms, as well optimizing the many math functions, filters, transforms, neural net libraries, and other software libraries, as part of a robust and long-term software ecosystem.
They've now been selected to provide the core CPU for NASA's next generation High-Performance Spaceflight Computing processor (or HSPC), according to a SiFive announcement: HPSC is expected to be used in virtually every future space mission, from planetary exploration to lunar and Mars surface missions.
HPSC will utilize an 8-core, SiFive® Intelligence X280 RISC-V vector core, as well as four additional SiFive RISC-V cores, to deliver 100x the computational capability of today's space computers. This massive increase in computing performance will help usher in new possibilities for a variety of mission elements such as autonomous rovers, vision processing, space flight, guidance systems, communications, and other applications....
The SiFive X280 is a multi-core capable RISC-V processor with vector extensions and SiFive Intelligence Extensions and is optimized for AI/ML compute at the edge. The X280 is ideal for applications requiring high-throughput, single-thread performance while under significant power constraints. The X280 has demonstrated a 100x increase in compute capabilities compared to today's space computers..
In scientific and space workloads, the X280 provides several orders of magnitude improvement compared to competitive CPU solutions.
A business development executive at SiFive says their X280 core "demonstrates orders of magnitude performance gains over competing processor technology," adding that the company's IP "allows NASA to take advantage of the support, flexibility, and long-term viability of the fast-growing global RISC-V ecosystem.
"We've always said that with SiFive the future has no limits, and we're excited to see the impact of our innovations extend well beyond our planet."
And their announcement stresses that open hardware is a win for everybody: The open and collaborative nature of RISC-V will allow the broad academic and scientific software development community to contribute and develop scientific applications and algorithms, as well optimizing the many math functions, filters, transforms, neural net libraries, and other software libraries, as part of a robust and long-term software ecosystem.
Easy to modify? (Score:5, Insightful)
Given how specific and extreme are the demands for space ready computers, i imagine they decided on it by the fact they can freely alter and customize the piece to get to their needs, and maybe even easier to make it radiation proof, that is the main reason why they still use old chips.
Re: (Score:2)
the fact they can freely alter and customize the piece to get to their needs
Indeed. RISC-V has an open license, but is also architecturally designed to be customized. It is easy to add or delete entire subsystems from the design. Don't need vector processing? Just set a flag to omit it.
even easier to make it radiation proof
Indeed. The most reliable component is the one that isn't there.
That is the main reason why they still use old chips.
The ole' reliable standard for rad-hard systems is the venerable 8051, an 8-bit design from 1980. Let's hope that RISC-V can finally replace it.
Re: Easy to modify? (Score:2)
Re: Easy to modify? (Score:5, Interesting)
Requirements vary a lot by mission. If its one of many satellites in a cluster in low earth orbit, its not so bad if one fails. If its a pluto mission, and if this part fails, it represents a >20 year delay before another launch, its a different matter. Or if it takes out a >10B space telescope.
Silicon Chip (Score:2)
Re: (Score:3)
The article is here [siliconchip.com.au] for those interested. However, it is $10 to purchase.
This looks to be an amazing magazine. Thanks!
Re: (Score:3)
It seems they have space certified / rad hardended chips at 12nm
https://spacenews.com/bae-syst... [spacenews.com]
Although that link above doesn't state what architecture those chips are in.
And the link in the summary above does not state what nm will the new chips be made in.
Re: (Score:2)
Why are space chips custom rather than shielded?
Most rad-hard chips are not custom. You can buy them off-the-shelf from Digikey and many other distributors.
They aren't just used in space. They are also used in applications where an SEU can cause a loss of life, such as aviation and defense.
Re: (Score:2)
The standard X280 CPU core has 32 vector registers of 512 bits each, for AI/NN and other processing.
That's the reason (along with the multiple cores) that it is stated to be 100x faster than the previous PowerPC "G3" RAD750.
Re: (Score:2)
Why are space chips custom rather than shielded?
Mass. Stopping cosmic rays is hard. You could need literally tons of shielding to make a significant difference.
Re: (Score:2)
Why are space chips custom rather than shielded? Is this CPU competitive to todays ARM and x86 CPUs? What about GPU capabilities for AI/NN applications like guidance based on visual inputs?
I too would love to be able to use an M1, but it's not space-rated.
Re: (Score:2)
Why are space chips custom rather than shielded?
Practical shielding intercepts cosmic rays resulting in a shower of particles which would be even worse. It is much more effective to use a semiconductor process, like silicon-on-insulator resulting in dielectric isolation, which is inherently radiation resistant.
Re: (Score:2)
Re: (Score:3)
Also, to be failsafe and highly redundant, it could be nice to use at least 2 completely different architectures to do the same computations.
Unlikely to matter. (Score:4, Interesting)
The industry will still exist, probably, and just push to a more ambitious limit, but it won't be the default by any stretch of the imagination.
Re:Unlikely to matter. (Score:5, Funny)
> That limitation is why you opt for custom hardware in space applications rather than just adding shield mass to more or less off-the-shelf technology.
It's pushing Intel micro-code updates to space 4 times a year that's an issue. :)
Re:Unlikely to matter. (Score:5, Insightful)
Re: (Score:3)
These new processors will be great. Certainly, the space sector could benefit from extra data processing capability, but we don't have 2 billion people all simultaneously streaming HD videos of Trump, porn, and Alex Jones.
Re: (Score:2)
Why not? The dimensions and mass performance would seem to easily allow it, and it would still be a lot cheaper and better-performing than anything that exists now.
Re: (Score:3)
It would add 50 to 100 kg to the mass. Would you rather use the 50 to 100kg for sensors/instruments, or for shielding an RTX 3090? A space probe just needs enough computing to run communications, spacecraft nav/orientation (less computing than a pocket calculator), basic non-realtime computational photography algorithms, and storage. We don't need to enable aliens to play Elden Ring on the GPU. By the way, contrary to popular belief, the 1970s-era Voyager space probes didn't even have a CPU and they still n
Re: (Score:3)
It would add 50 to 100 kg to the mass.
I think you dropped an order of magnitude if you calculate that "2 feet of water or concrete around each PC" would adds only 50 to 100 kg. You have to shield it from all directions.
Re: (Score:2)
Hmm .. you're right.. wow that's mad.
Re:Unlikely to matter. (Score:5, Funny)
2 billion people all simultaneously streaming HD porn videos of Trump and Alex Jones.
Dyslexia is a nightmare.
"Heck of a lot" is something Starship can do. (Score:2)
Re: (Score:2)
The specs are mind-boggling when you start thinking about implications for payload development.
The amount of shielding to block galactic cosmic rays is mind-boggling when you start thinking about the implication of twenty tons of shielding to protect 0.1 grams of silicon.
Re: (Score:2)
Think of it like this: Which do you think would be cheaper and sell more... a phone developed with specialty electronics to function underwater, or a waterproof case for regular phones? So far, the limits of orbital rockets have forced the satellite industry to choose the former approach eve
Re: (Score:2)
I'm not sure where "20 tons" is coming from. A cubic meter of water is only 1 tonne, and you don't need anywhere near that to keep the operating environment within range of straightforward error-correction augments.
A meter of water might be good enough shielding against solar proton events, but the post you're responding to said "galactic cosmic rays", and you'll need more than that to stop GCRs. Two meters of water surrounding a chip would be 4/3pi r^3=33 tons, so that's an order of magnitude more than the "only 1 tonne" you estimate.
The worst case would be GCR leaving an ionization track behind it that creates a short between the high voltage and ground, and the short latching the circuit up so that it doesn't turn
Re: (Score:2)
Re: (Score:2)
Re:Unlikely to matter. (Score:5, Informative)
"Space computing" is just about rad hardening
Rad-hardening is one issue. Power consumption is another. More power means bigger solar panels and bigger batteries, which boosts launch costs. More power also means more heat, and shedding heat is difficult in a vacuum.
The biggest mitigation for radiation is to use depleted boron [wikipedia.org] as a semiconductor dopant. SOI (silicon-on-insulator) fabrication also helps. SRAM is more resilient than DRAM, so should be used for any critical memory that isn't scrubbed. None of these techniques are expensive, and all are way cheaper than more shielding.
Re: (Score:2)
True, but the same logic applies to power systems: The added cost is trivial if launch costs are $100k per ton to LEO (which is 10x more expensive than Starship is actually aiming for) with a cavernous 1000 cubic-meter volume limit.
Re: (Score:2)
Economizing still means more birds in one launch. It does ultimately come down to a cost decision. A single rad-hardened CPU has historically cost hundreds of thousands of dollars, so it really depends on what these new units will cost.
Re: (Score:2)
SRAM is more resilient than DRAM, so should be used for any critical memory that isn't scrubbed.
Discrete SRAM is more resistant, but the integrated SRAM used for caches is at least an order of magnitude worse than DRAM, which is why it has included ECC or parity protection from almost the time that it was first used.
Re: Unlikely to matter. (Score:1)
Competing with off the shelf "custom" hardware and software standards was never a thing... As well as competing with a private commercial program will not be a thing...
Re:Unlikely to matter. (Score:4, Interesting)
That limitation is why you opt for custom hardware in space applications rather than just adding shield mass to more or less off-the-shelf technology. But Falcon 9 is already loosening that limit, and Starship will sooner or later obliterate it completely.
It really won't.
There's a very simple calculation you can do: look at the atmospheric pressure (10^5) Pa, and divide by g, roughly 10m/s/s. For a meter square that is 10 tons of stuff above you to shield you from radiation. For water that would be 10m of water. For lead, to get the equivalent shielding from all sides, you'd need to sit inside a ball of radius 1m or so (very roughly). Lead's a bit denser than 10, and it's absorption coefficient is different from air, but not by a colossal amount, see e.g.:
Nitrogen https://physics.nist.gov/PhysR... [nist.gov]
Lead https://physics.nist.gov/PhysR... [nist.gov]
the precise numbers are hugely complex and depend on the type of radiation particles and the energy spectrum of the radiation and of course the precise composition, and that only applies to x rays. But they still aren't that different.
A 1m radius sphere of lead weighs about 40 tons. My back of envelope calculations might be wrong. If I'm out by a factor of two in the optimistic direction, you're still looking at 5 tons of lead to shield the electronics as much as the atmosphere does. And you still need a sphere a meter across!
Re: (Score:2)
But Falcon 9 is already loosening that limit, and Starship will sooner or later obliterate it completely. It will be a trivial cost to just encrust OTS electronics in extra shielding than to keep manufacturing these extremely rare systems.
BUT...Mass is not only a problem at launch. It's a problem anywhere in the lifetime of a mission when thrust has to be applied to change course. The more mass you have, the more fuel and propellant you have to bring along on the mission.
SiFiveî (Score:1)
I think the editor needs to edit the summary.
Re: (Score:1)
What's with you Q-tards and the NASA hate anyway? Is it spillover from when you spent your days babbling about how the moon landing was faked?
Re: (Score:3)
Anybody else can make a RISC-V without licensing fees, so there's nothing preventing them from seeking another supplier for an equivalent part. They're planning a part that has plenty of capacity, and if some of it's unused then it can simply be disabled and kept in reserve for future hack value. One platform and one development system means a lot less incompatibility.
So it's official now... (Score:5, Funny)
... space travel is not without RISC.
Re:So it's official now... (Score:4, Informative)
Nice one!
RISC-V won't be the first RISC taken in space... Ingenuity uses an ARM processor running Linux. Pretty cool stuff!
Re:So it's official now... (Score:5, Informative)
pssst [wikipedia.org]
Re: (Score:2)
That's true. NASA has been taking RISCs for years!
However it's still pretty cool a consumer cellphone ARM CPU running Linux has manage to survive for so long in such a harsh environment. Heard a couple of interviews with one of the project leads and it was very interesting. Including shell access and python scripts, and lots of ffmpeg.
I'm sure we'll see more use of cheap, off-the-shelf hardware for demonstrations in the future. However, no doubt the next mars copter will use actual hardened components (pe
Re: So it's official now... (Score:4, Informative)
The SPARC ERC32, MIPS Mongoose V, and the RAD6000 all predate the RAD750. The space going SPARC line continues on with the LEON series.
Anything (Score:2)
Why... (Score:3, Funny)
Re: (Score:2)
Space Computers, huh? (Score:2)
"The X280 has demonstrated a 100x increase in compute capabilities compared to today's space computers.."
*laughs in space alien*
Let us take a moment (Score:2)
On how the Space Shuttle flew on 386 level of processor technology by the time it was retired, and the Apollo missions used a computer that was less powerful than a scientific calculator you can buy at the dollar store.
Compute literacy (Score:2)
"Compute" is a verb. Stop using it as a noun. It makes one look illiterate.
Thankfully T2 Linux has 1st class RISCV support (Score:2)