Sony's ReRAM plans - what can and can't ReRAM bring to a console? *spawn

[megre]

Sony will commercialize Reram in 2020.

7:34:05 - Start of Sony Presentation

7:42:40 - 8 chips of 128gbit = 128Gbytes, 25.6GB/s read, 9.6GB/s write;
16 chips of 128gbit = 256Gbytes, 51.2GB/s read, 19.2GB/s write;
8 chips of 64gbit = 64GB, 25.6GB/s read, 9.6GB/s write;
4 chips of 128bit = 64GB, 12.8GB/s read, 4.8GB/s write

7:45:15 - "We accelerated to 2020"

8:09:30 - Why nand-based SSD has no future. "Nand flash stop scaling at 15nm". Reram-based SSDs have the potential to be cheaper in the future.

Sony emphasizes that its ReRAM has a smaller power consumption and can be more easily densified than the cross point-type phase-change memory (PCM) used by the Optane.

There's a lot at stake here for Sony. It could be a new line of business for them. This should be expensive at first as with all technology because of yields but once it matures it should not be much more expensive than nand chips to produce. I could see Sony eating up production costs initially so they can position its technology.

 

[KeanuReeves]

Am I understanding this correctly that they are preparing to enter the storage market with this technology? I read this article from (warning)wccftech where Samsung spoke about SSD's for next-gen consoles and I assumed Sony would go with them. Or am I just completely lost?

At the recent Samsung SSD Forum 2019 held in Tokyo, as reported by PC Watch, Senior Managing Director, Product Planning Team, Memory Electronics Division Han Jinman spoke about the company's storage solutions. At one point, gaming SSD has been mentioned, with the accompanying slide making it clear how optimized NVMe SSD will reduce both system booting times and loading times, confirming that they will be installed in both the PlayStation 5 and Xbox One Scarlett.

 

[Shifty Geezer]

Am I understanding this correctly that they are preparing to enter the storage market with this technology? I read this article from (warning)wccftech where Samsung spoke about SSD's for next-gen consoles and I assumed Sony would go with them. Or am I just completely lost?

The way that's worded I think they are saying the slides confirmed inclusion, but a cursory glance at the slides makes them look illustrative that Samsung's SSDs could benefit consoles. That 'confirmation' seems to come from wccftech and not Samsung.

 

[Silent_Buddha]

Am I understanding this correctly that they are preparing to enter the storage market with this technology? I read this article from (warning)wccftech where Samsung spoke about SSD's for next-gen consoles and I assumed Sony would go with them. Or am I just completely lost?

The Sony presentation doesn't necessarily mean the PS5 will be using ReRAM although the 2020 date could hint at its use in the console.

At the moment, IMO, it's more likely that the PS5 will be using NAND based solid state storage as ReRAM is likely to be far to expensive to include in a less than 500 USD console. Of course there is always the possibility that Sony want to sell the PS5 at a massive loss in order to try to push the tech. I'm doubtful of that, however.

While it may eventually be cheaper than NAND, it's just a potential for some nebulous point in the future.

Regards,
SB

 

[MrFox]

Sony will commercialize Reram in 2020.

7:34:05 - Start of Sony Presentation

7:42:40 - 8 chips of 128gbit = 128Gbytes, 25.6GB/s read, 9.6GB/s write;
16 chips of 128gbit = 256Gbytes, 51.2GB/s read, 19.2GB/s write;
8 chips of 64gbit = 64GB, 25.6GB/s read, 9.6GB/s write;
4 chips of 128bit = 64GB, 12.8GB/s read, 4.8GB/s write

7:45:15 - "We accelerated to 2020"

8:09:30 - Why nand-based SSD has no future. "Nand flash stop scaling at 15nm". Reram-based SSDs have the potential to be cheaper in the future.

Sony emphasizes that its ReRAM has a smaller power consumption and can be more easily densified than the cross point-type phase-change memory (PCM) used by the Optane.

There's a lot at stake here for Sony. It could be a new line of business for them. This should be expensive at first as with all technology because of yields but once it matures it should not be much more expensive than nand chips to produce. I could see Sony eating up production costs initially so they can position its technology.

I'm pessimistic about this. Many companies, including Sony, have developped an ReRAM, or MRAM, or PCRAM cell structure for years, but they need a memory-class fab partner. Without that scale of production it will never be as inexpensive as nand. Right now they all cost a fortune.

Sony is a semiconductor company with very specialized fabs (true 3d stacks, millions of cu-pillars, best sensors tech), but they are not a commodity memory scale producer. They only do high end high margin silicon.

Nand is dirt cheap, there's extreme competition, and gigantic volumes.

This is 2 years old but still valid I think....
https://semiengineering.com/what-happened-to-reram/

 

[megre]

I'll just post some interesting non-technical slides from the sony presentation from 2017 and 2019.

 

[megre]

Many companies, including Sony, have developped an ReRAM, or MRAM, or PCRAM cell structure for years, but they need a memory-class fab partner.

Sony is a semiconductor company with very specialized fabs (true 3d stacks, millions of cu-pillars, best sensors tech), but they are not a commodity memory scale producer.

I'm not sure about that. https://pc.watch.impress.co.jp/docs/column/semicon/1123505.html (translated from japanese)

Avalanche Technology has commercialized "pSTT-MRAM", which combines perpendicular magnetic recording magnetic tunnel junction (pMTJ) and spin injection. According to the company's website, the storage capacity of the product is not so large, 4Mbit / 8Mbit. Assume that 64Mbit large-capacity products are being sampled.

　The talk stated that the pSTT-MRAM product with 55nm / 40nm technology is in mass production. It is interesting that Sony is the silicon foundry. He also explained that 256Mbit pSTT-MRAM using 2Xnm technology is currently under development and will be commercialized in 2019.

The link also discusses sony's reram technology in detail.

 

[Shifty Geezer]

What's 'mass production' though? How many people are using pSTT-MRAM such that mass production is needed?

 

[megre]

What's 'mass production' though? How many people are using pSTT-MRAM such that mass production is needed?

Good point. I don't know. It's translated from Japanese and I don't believe it's mass production per se. But it looks like Sony has the capacity to produce them memories.

Didn't they produce their proprietary memory stick though? I think Sony has the technology and engineers.

 

[Shifty Geezer]

Well, MrFox says they have fabrication technology. They may well be making specialist, high-margin chips but these would likely be too expensive to include in PS5 versus using NAND. There needs to be a good argument for using a non-commodity tech over cheap-as-chips solutions, such as Sony already mass-producing these things for various products, or investing heavily in the tech on the assumption they'll be in demand.

 

[Entropy]

Well, MrFox says they have fabrication technology. They may well be making specialist, high-margin chips but these would likely be too expensive to include in PS5 versus using NAND. There needs to be a good argument for using a non-commodity tech over cheap-as-chips solutions, such as Sony already mass-producing these things for various products, or investing heavily in the tech on the assumption they'll be in demand.

Well actually he says that they don’t, and would need to collaborate with a memory manufacturer.

I'm pessimistic about this. Many companies, including Sony, have developped an ReRAM, or MRAM, or PCRAM cell structure for years, but they need a memory-class fab partner. Without that scale of production it will never be as inexpensive as nand. Right now they all cost a fortune.

Sony is a semiconductor company with very specialized fabs (true 3d stacks, millions of cu-pillars, best sensors tech), but they are not a commodity memory scale producer.

Doesn’t change anything really, otherwise. Actually, collaboration would vastly increase the odds of the technique materializing.
Sony has access to 45/40 nm fabs, but their next fab investment targets image sensors, not memory.

 

[Silent_Buddha]

Well actually he says that they don’t, and would need to collaborate with a memory manufacturer.

That's not what he said.

...but they are not a commodity memory scale producer.

Doesn't mean that they don't produce any, just that don't produce it in large enough quantities to compete on the commodity memory market. So, it's entirely possible that they make specialized products in that segment, but since it isn't sold in the commodity market, the prices are going to be rather high.

Regards,
SB

 

[MrFox]

They need a memory fab partner if they want to reach competitive pricing against nand products. They can certainly make niche high priced parts themselves. (and they do)

Interestingly, sony did partner with micron for eventually making their reram, until micron decided to drop reram in favor of phase change and partner with intel for optane. The fact that even intel partners with a memory fab seems to validate what I mean by gigantic economies of scale required to compete in the commodity memory business.

TSMC can make reram, but it's crazy to think that would compete in price against the likes of samsung or micron.

Integrating say 64GB of reram might be possible on ps5, at the cost of a 1TB nvme pcie4, so it's a weird proposition.

 

[megre]

investing heavily in the tech on the assumption they'll be in demand.

https://www.networkworld.com/article/3453380/5-disruptive-storage-technologies-for-2020.html

SCM is not merely faster than NAND-based flash alternatives—it's in the range of 1,000-times faster. "Microsecond latency, not millisecond," Watson says. "It’s going to take some time to wrap our collective heads around what this will mean for our applications and our infrastructure," he adds. SCM's initial big play will be extending memory, Watson predicts, noting that third-party software already allows in-memory applications to use Optane to achieve footprints of up to 768TB.

The fact that data centers planning to adopt SCM will be restricted to deployment on servers using the latest-generation Intel CPUs (Cascade Lake) threatens to mute the technology's immediate impact. "But the ROI may turn out to be so irresistible that it could drive a wave of data center upgrades to embrace the unfolding opportunities associated with this major sea change," Watson says.

https://venturebeat.com/2019/11/13/...ologies-that-will-change-how-you-handle-data/

Coughlin Associates predicts that 3D XPoint memory—the technology at the heart of Optane—will drive revenues to over $16 billion by 2028. Clearly, there’s a demand for new memories that address the impending limits of flash memory, DRAM, and SRAM.

Intel’s Optane DC persistent memory technology, which bridges the gap between DRAM and SSDs, magneto-resistive RAM with ambitions to excel in edge computing applications, and ReRAM, a potential replacement for flash memory.

http://hosted-p0.vresp.com/1056991/f253129da6/ARCHIVE

Emerging Memories such as MRAM and RRAM will soon offer huge opportunities for users and vendors
Produced by highly respected analysts Tom Coughlin and Jim Handy, the new report projects a $20 billion market by 2029.

 

[megre]

Remember Optane is exclusive to Intel CPUs. (edit: i googled and it seems that it's not true anymore?)

How will AMD compete in the cloud enterprise architecture?

 

[itsmydamnation]

Remember Optane is exclusive to Intel CPUs. (edit: i googled and it seems that it's not true anymore?)

How will AMD compete in the cloud enterprise architecture?

By offering more higher performance cores with more memory bandwidth while being cheaper then the competition. Just look at all the super computer announcements as an example. The project I'm working on now will buy EPYC in h1 next year.

 

[megre]

By offering more higher performance cores with more memory bandwidth while being cheaper then the competition. Just look at all the super computer announcements as an example. The project I'm working on now will buy EPYC in h1 next year.

It looks like the project your working on doesn't benefit from ultra-fast persistent memory like optane but for those where a persistent memory like optane give a huge benefit, how will amd compete?

It seems like those who need fast persistent memory has no choice but to go intel for now. Is that why stadia has amd gpu but intel cpu?

 

[itsmydamnation]

It looks like the project your working on doesn't benefit from ultra-fast persistent memory like optane but for those where a persistent memory like optane give a huge benefit, how will amd compete?

It seems like those who need fast persistent memory has no choice but to go intel for now. Is that why stadia has amd gpu but intel cpu?

Yep welcome to the real world, vast majority of workloads won't be using persistant memory just like they still don't support simd and if they do it will be sse, #enterpriseshitwareforlife

 

[Shifty Geezer]

https://www.networkworld.com/article/3453380/5-disruptive-storage-technologies-for-2020.html
https://venturebeat.com/2019/11/13/...ologies-that-will-change-how-you-handle-data/
http://hosted-p0.vresp.com/1056991/f253129da6/ARCHIVE

These may show there's demand, but is there evidence of Sony investing in it? We should have announcements of Sony buying/building a fab for the purpose, same way we've heard about them building fabs for other semiconductor ventures. If Sony doesn't have the resources in place to mass produce this stuff for PS5 (and everyone else), it ain't going in PS5.

 

[megre]

We should have announcements of Sony buying/building a fab for the purpose, same way we've heard about them building fabs for other semiconductor ventures.

I understand that. And I've been data mining looking for clues. I was actually the one who found the video with the sony presentation. Right now to me it seems like: a. ps5 reram is not happening, b. sony has repurposed one of their fab to manufacture it themselves, c. sony has a fab partner

Should we have announcements about it though? Is that a given? Is there no point for Sony to be secretive about it you think?