MIPS is Back

[Sal Ammoniac]

Wave Computing, the company that currently owns the MIPS architecture, and which declared chapter 11 bankruptcy in April 2020, is emerging from bankruptcy and rebranding itself as MIPS.

As part of the announcement, they also said that they're developing a new "standards-based 8th generation architecture" which will be based on the RISC-V processor standard. So MIPS is back. Sort of.

https://eetimes.com/wave-goodbye-hello-mips-as-chapter-11-resolved/

[SiliconWizard]

Am I the only one hesitating between hysterical laugh and facepalming thinking about MIPS ever since it got acquired by Wave Computing? What a disaster...

[Sal Ammoniac]

I don't have much hope of them being able to rescue the MIPS architecture from obscurity, not with ARM and RISC-V as their main competitors. I suppose that's why they're planning on moving to RISC-V themselves for their next-gen core.

[james_s]

I remember MIPS from the SGI machines. Never heard of Wave though.

[SiliconWizard]

I don't have much hope of them being able to rescue the MIPS architecture from obscurity, not with ARM and RISC-V as their main competitors. I suppose that's why they're planning on moving to RISC-V themselves for their next-gen core.

That was certainly a challenge, but now the story seems basically over. Except if "MIPS" gets once again in other hands - which very well may be. Is MIPS becoming a soap opera?

What would be nice is if they at last open sourced their MIPS ISAs for good (especially the latest, that I think have never really seen the light in actual processors?) if they are switching to RISC-V. I don't hold my breath though. I'm sensing the saga is going to continue.

[coppice]

What would be nice is if they at last open sourced their MIPS ISAs for good (especially the latest, that I think have never really seen the light in actual processors?) if they are switching to RISC-V. I don't hold my breath though. I'm sensing the saga is going to continue.

That might be something they can't do. Some of their architectures are still in active use, and there will be agreements with existing licensees.

[Sal Ammoniac]

I remember MIPS from the SGI machines. Never heard of Wave though.

Neither had I. I still use MIPS in the form of the PIC32.

[SiliconWizard]

You probably missed the acquisition of MIPS by Wave Computing, and the associated MIPSopen initiative.
There were two threads about this on this forum (maybe others?):

https://www.eevblog.com/forum/microcontrollers/mips-goes-open-source
https://www.eevblog.com/forum/oshw/mips-open-open-licence

The beginning: https://www.forbes.com/sites/moorinsights/2018/06/13/ai-pioneer-wave-computing-acquires-mips-technologies/
The "end": https://www.reuters.com/article/wavecomputing-mips/ai-chip-firm-wave-computing-emerges-from-chapter-11-bankruptcy-protection-idUSL2N2KZ33N

Frankly, when I saw some AI startup acquiring MIPS, I knew it wasn't going to end well.

[amyk]

RISC-V is basically MIPS, more or less.

[brucehoult]

RISC-V is basically MIPS, more or less.

RISC-V is similar to MIPS ISAs, especially the more recent ones which came after RISC-V already existed such as NanoMIPS, and MIPS R6 in 2014 that added things such as conditional branches comparing two registers which RISC-V already had.

The instruction mnemonics are quite similar, but RISC-V has completely different binary encoding with a lot more opcode space available for future extensions, brought about largely by the RISC-V 20-12 split of LUI vs the immediate/offset on all the arithmetic and load/store instructions vs 16-16 for MIPS.

[rcbuck]

Will this have any effect on the Microchip PIC32 parts? Or nothing changes since they have already paid for licensing?

[brucehoult]

Will this have any effect on the Microchip PIC32 parts? Or nothing changes since they have already paid for licensing?

Of course customers who are already making stuff can keep making it.

I'm sure MIPS would sell new customers MIPS-ISA cores if they want them. Why not? It is of course the lack of such customers that sent them bankrupt in the first place.

What might be interesting is whether this pushes Microchip to adopt more RISC-V more quickly, not just supporting a RISC-V soft core in their FPGAs and RISC-V cores in their PoplarFire SoC FPGAs.

[martinribelotta]

What might be interesting is whether this pushes Microchip to adopt more RISC-V more quickly, not just supporting a RISC-V soft core in their FPGAs and RISC-V cores in their PoplarFire SoC FPGAs.

The hope is shared by my but the direction apears diferent:
https://www.microchip.com/en-us/products/microcontrollers-and-microprocessors/32-bit-mcus/pic32-32-bit-mcus/pic32cm-mc

In my opinion the correct mchip move should be embrace RISC-V ISA and populate heavy processors with this like NXP's crossover processors... But I am not the CTO of mchip 

[SiliconWizard]

What might be interesting is whether this pushes Microchip to adopt more RISC-V more quickly, not just supporting a RISC-V soft core in their FPGAs and RISC-V cores in their PoplarFire SoC FPGAs.

The hope is shared by my but the direction apears diferent:
https://www.microchip.com/en-us/products/microcontrollers-and-microprocessors/32-bit-mcus/pic32-32-bit-mcus/pic32cm-mc

In my opinion the correct mchip move should be embrace RISC-V ISA and populate heavy processors with this like NXP's crossover processors... But I am not the CTO of mchip 

Also see the other thread about Microchip. At this point, it's very hard to tell what they will do, but even more humbly, it's hard to tell what the "right" direction would be. Honestly.

RISC-V is all nice, but, and it looks like we have to say it again and again, it's NOT hardware. RISC-V is just a bunch of specs. Ever since the PIC32, Microchip has apparently decided to stop designing their own cores. The MIPS cores they used were pretty much "ready to use" with the right support, as is the case with ARM.

So you would have to define what going RISC-V would mean. I doubt they'd be ready to go back to designing their own cores, even if now RISC-V-based, whereas they have pretty much stopped doing this and this approach is becoming almost non-existent in the MCU industry. When deciding to make ARM-based, or MIPS-based MCUs, a company doesn't just select an ISA. Actually, the ISA itself may not matter all that much as long as it fits their basic requirements. A company also goes for a vendor that will license ready-to-use (or almost) cores and all the tech support required to make a chip with them. So if they are going RISC-V, they will in all certainty do the same thing, and will have to select a vendor selling licenses of RISC-V cores. There are a few out there, but none has the history of either ARM or MIPS, so this choice alone will be a very tough one to do. Now since we've heard that the new MIPS would actually be designing RISC-V cores, that might be what happens in the end: Microchip again contracting cores with MIPS!

But at this point, and given the history, I would be pretty wary of MIPS' future myself, and would probably think twice before ever deciding to do business with them. Sorry MIPS but yeah...

[JPortici]

Ever since the PIC32, Microchip has apparently decided to stop designing their own cores. The MIPS cores they used were pretty much "ready to use" with the right support, as is the case with ARM.

Two years ago the dsPIC33C came out to the public, giving more Oomph in the form of some new handy instructions (bitfield manipulation, 6 cycle division, multiple contexts of all registers that can be swapped in a couple of cycles) higher max MIPS and dual core options

dsPIC33A is also in the works. dsPIC33A is a 32bit version of the dsPIC. Hopefully with fast bus peripherals (high speed USB, ethernet, camera interface) that were never part of the dsPIC line. When we will be able to see that? i don't know. I only know it exists (i was told about it, but it can't be a secret anymore, there is a dsPIC33A entry in the MPLABX IPE)

[PCB.Wiz]

Two years ago the dsPIC33C came out to the public, giving more Oomph in the form of some new handy instructions (bitfield manipulation, 6 cycle division, multiple contexts of all registers that can be swapped in a couple of cycles) higher max MIPS and dual core options

dsPIC33A is also in the works. dsPIC33A is a 32bit version of the dsPIC. Hopefully with fast bus peripherals (high speed USB, ethernet, camera interface) that were never part of the dsPIC line. When we will be able to see that? i don't know. I only know it exists (i was told about it, but it can't be a secret anymore, there is a dsPIC33A entry in the MPLABX IPE)

Interesting, and that makes some sense, as TI are expanding their DSP's ( C2000 real-time microcontrollers )

https://www.ti.com/microcontrollers-mcus-processors/microcontrollers/c2000-real-time-control-mcus/products.html#&sort=p1130;asc

Latest ones there are 100MHz with high precision PWM and capture.

[brucehoult]

So you would have to define what going RISC-V would mean. I doubt they'd be ready to go back to designing their own cores, even if now RISC-V-based, whereas they have pretty much stopped doing this and this approach is becoming almost non-existent in the MCU industry. When deciding to make ARM-based, or MIPS-based MCUs, a company doesn't just select an ISA. Actually, the ISA itself may not matter all that much as long as it fits their basic requirements. A company also goes for a vendor that will license ready-to-use (or almost) cores and all the tech support required to make a chip with them. So if they are going RISC-V, they will in all certainty do the same thing, and will have to select a vendor selling licenses of RISC-V cores.

Microchip already licenses both RISC-V soft cores for PolarFire FPGAs and hard cores for PolarFire SoC from SiFive, so that's a pretty obvious candidate.

The new RISC-V based MIPS is also obviously someone they already have a business relationship with.


I don't know of anyone except ARM that you can license a microcontroller-class ARM core from -- as far as I know, those such as Samsung and Qualcomm with ARM Architecture3 Licenses who sell ARM cores only have application processor cores -- and I think they only sell complete SoCs, not cores.

[westfw]

I don't know of anyone except ARM that you can license a microcontroller-class ARM core from

Well, of course not.  ARM *owns* ARM; I'd expect that sub-licenses are explicitly prohibited by their license terms.There are some other "cores" out there.  Tensilica (as used in ESP8266 and ESP32) and ARC (Intel Curie, however ill-fated) have appeared in microcontroller-like implementations.  And while RISC-V may be "only" an ISA specification, there are already companies that will sell you a full core implementation (SiFive, etc)

[brucehoult]

I don't know of anyone except ARM that you can license a microcontroller-class ARM core from

Well, of course not.  ARM *owns* ARM; I'd expect that sub-licenses are explicitly prohibited by their license terms.There are some other "cores" out there.  Tensilica (as used in ESP8266 and ESP32) and ARC (Intel Curie, however ill-fated) have appeared in microcontroller-like implementations.  And while RISC-V may be "only" an ISA specification, there are already companies that will sell you a full core implementation (SiFive, etc)

There are far more commercial, supported, RISC-V cores that just the ones from SiFive!

For a start, there is Andes, which is bigger and does more business than SiFive.  Then there is at least IQonIC Works, Syntacore, Codasip, T-Head (Alibaba), CloudBear, Nuclei. Plus there are open-source but commercially supported cores such as from Cobham Gaisler and PULP project.

[westfw]

There are [many] commercial, supported, RISC-V cores

Do you know offhand how many are doing "microcontroller-like" cores (say: 32bit only, no MMU, not linux-capable.)I looked at the list on the , but the size and options covered are seriously in need of some sort of parametric search...

[EEVblog]

Will this have any effect on the Microchip PIC32 parts? Or nothing changes since they have already paid for licensing?

I was wondering that. Surely they would have secured an arrangement that allows then to use the IP indefinitely. If MIPS went bust that was their problem.

[brucehoult]

There are [many] commercial, supported, RISC-V cores

Do you know offhand how many are doing "microcontroller-like" cores (say: 32bit only, no MMU, not linux-capable.)I looked at the list on the , but the size and options covered are seriously in need of some sort of parametric search...

Virtually all of them.

Anything 64 bit, or with MMU, or even with FPU is very much the exception.

I can't think of anyone who is doing only Linux-capable and no microcontrollers -- except maybe Alibaba, but I'm not sure about that.

[amyk]

and ARC (Intel Curie, however ill-fated)

Wasn't that a 486 on a smaller process? The only use of ARC I've heard of in Intel products is their wireless cards and the Management Engine.

[westfw]

Curie had a small x86, but also an ARC co-processor. (And a dsp of some kind, and a neural net thing.). The arduino 101 core ended up compiling code for the arc, while the x86 did “other stuff.”   (Yeah, that always seemed a bit weird.)

[martinribelotta]

And then, the MIPS (architecture) is dead, long live to MIPS (provider) that not provides MIPS ISA, only RISC-V:

https://www.eejournal.com/article/wait-what-mips-becomes-risc-v/