Print 37 comment(s) - last by mik123.. on Aug 21 at 7:55 PM

  (Source: Crossbar)
Startup claims gaudy performance numbers -- but can it deliver

Even with the small size, high speeds, and power efficiency of NAND flash; modern mobile devices -- from wearable electronics to smartphones -- are demanding even better performance.  In the race to find a successor to flash many are looking at resistive random access memory (ReRAM/RRAM), a new form non-volatile memory (NVM) as a potential "front runner" in the race.

I. Crossbar Emerges From the Shadows, Passes Hynix in RRAM Race

Hewlett-Packard Comp. (HPQ) seemed to get the jump in the ReRAM race when it became the first to implement what electronics experts dub a "memristor" in mid-2008.  As of 2010 HP was still tweaking the technology, looking to push down latency to commercially competitive levels.  Independent researchers in the mean time were tweaking flexible and transparent memristors. In Oct. 2011, HP vowed to release a memristor-powered SSD "within 18 months" -- but 22 months later it has no commercial product.

Stan Williams, an HP Labs senior fellow, revealed late last year that no memristor product was in store for 2013, and that HP was offloading the technology to its partner SK Hynix Inc. (KRX:000660).  He explains, "In terms of commercialization, we will have something technologically viable by the end of next year. Our partner, Hynix, is a major producer of flash memory, and memristors will cannibalize its existing business by replacing some flash memory with a different technology. So the way we time the introduction of memristors turns out to be important."

Perhaps the company's crippling, historic layoffs had something to do with that pullout.  But whatever the reason, tech observers that had waited and hoped for a commercial RRAM product were left staring a launch slipping to sometime in 2015 or beyond (via Hynix).

Crossbar is the first to show off full, taped out RRAM test chips.  And its early product is significantly superior to flash.

But in a surprise the world now has its first commercial-ready ReRAM product -- and it’s not from HP/Hynix.  The new device is dubbed the "Crossbar Array" and comes from a virtually unknown startup named "Crossbar, Inc.".  

II.  Silicon's Best Kept Secret Was Born Out of the Fall of Spansion

Crossbar was co-founded by George Minassian.  With a Ph.D in electrical engineering and computer science from the Univ. of Texas at Austin, Mr. Minassian has been a quiet, but important player in the personal computing industry over the last two decades.

A fellow at Advanced Micro Devices, Inc. (AMD) for 19 years, he served as director of wireless during the golden "Athlon" era, assuming that post in 1999.  He led a team to produce AMD's first 802.11b/a chipset designs, as well as the designing a major RF fabrication process.

George Minassian
George Minassian, CEO of Crossbar, is no stranger to high technology.

In 2003, Mr. Minassian was shuffled by AMD from his post as wireless director to be VP of System and Software Engineering at memory maker Spansion Inc. (CODE).  When Spansion completed its 2008 spinoff via IPO, Mr. Minassian stayed onboard, serving largely in a public relations role at the chipmaker.

In 2009, Spansion was the world's third largest chipmaker, but was plagued by the global recession, falling NAND commodity prices, and a weak Q4 2008. The company entered into Chapter 11 bankruptcy in March 2009.  In Dec. 2010, more bad news arrived as Spansion lost a U.S. International Trade Commission (ITC) complaint against Samsung Electronics Comp., Ltd. (KSC:005930), which Spansion was hoping to squeeze for licensing fees.  But by then Mr. Minassian had already departed, off to return to his roots -- big engineering.

The fall of Spansion fueled the rise of Crossbar. [Image Source: Cap. Arch. Signs, Austin]

2010 seemed an unlikely year to give birth to what would become an aggressive startup -- but it was also a good year to leave Spansion.  Today Spansion has fallen to fifth in the memory market, even as it recovered.

But for Mr. Minassian, the decision to leave Spansion was not merely about the company's struggles.  He also had grown tired of being confined to what was increasingly a public relations desk job.  He was an engineer.

So later that year he launched Crossbar, based out of Santa Clara, Calif. (Crossbar was incorporated, but inactive from 2008-2009).  Over the next three years Mr. Minassian privately pitched his vision of the next big thing in the storage space -- resistive RAM (ReRAM).  Some investors bit.

III. Crossbar Peeks up Its Periscope With SEC Filing, Draws Short-Lived Buzz

In Nov. 2012, eyebrows were raised when the unknown company quietly announced in a mandatory U.S. Securities and Exchange Commission (SEC) filing that it had raised $20.5M USD in seed funding from Kleiner Perkins Caufield & Byers, Artiman Ventures, and Northern Light Venture Capital.  The University of Michigan was also a minority investor in the seed round.

The filing revealed that Crossbar was also being led by VP Hagop Nazarian -- a fellow refugee from Spansion who had served there as VP of Design Engineering.  Before that he had worked at Micron Technology, Inc. (MU), Information Storage Devices, Cypress Semiconductor Corp. (CY), and Xicor.

The filing also listed Anne Dorman as chief financial officer (CFO) for the firm.  She was a specialist in startups, and it 2012 was also listed as the CFO of Hornet Networks (a gay social network with 1m users), RelayRides (a car sharing service), Coraid (a cloud storage firm), Bandcamp (A SoundCloude-esque music discovery service), SuVolta (a CMOS power optimization firm), and Vii.

The Silicon Valley: Business Journal wrote:

Much less is known about Crossbar, which said in a Form D filing with the Securities and Exchange Commission that it has raised $20.5 million of a $21 million round.

The company didn't return phone calls asking about the new funding, but says on its website that it is "focused on delivering solutions that will dramatically enhance the global memory storage industry."

It isn't known if the company is named after cutting edge mobile storage architecture called crossbar, but it would seem logical. This technology breaks down the barrier between memory circuitry and logic circuitry, possibly providing more capacity in a smaller space.

Indeed, that was precisely what Crossbar -- which in total raised $25M USD in seed money -- was formulating.  But news of the mysterious startup quickly died down from Dec. 2012 until this month when -- to quote the company's press release -- "Crossbar emerge[d] from stealth mode."

IV. Dropping the Hammer

This week Crossbar flipped the switch on a new site and a press release that bore surprising news -- the company had a commercially read ReRAM device.  The company describes its prototype, which was successfully manufactured on an unnamed commercial third party fab, writing:

The Crossbar memory cell is based on three simple layers: A non-metallic bottom electrode, an amorphous silicon switching medium and a metallic top electrode. 

Crossbar's design is simple, with three layers.

The chip was manufactured on the 25 nm node.  Here's how the 64 Gigabit (8 GB) chip stacks up to its main rival, NAND flash chips of the same capacity at that node:
Mr. Minassian is bullish on his company's product, stating, "What is unique about this is that we have been able to get to manufacturing in just three years.  It is a technology that is easy to manufacture."

"Today’s non-volatile memory technologies are running out of steam, hitting significant barriers as they scale to smaller manufacturing processes.  With our working Crossbar array, we have achieved all the major technical milestones that prove our RRAM technology is easy to manufacture and ready for commercialization. It’s a watershed moment for the non-volatile memory industry."

Crossbar, stackable
Crossbar's cells are fully stackable, just like NAND offering an early route to dension expansion.

The technology is capable of supporting multiple levels of stacked cell/crossbar layers, plus it is fully compatible with CMOS integration.

Crossbar RRAM CMOS integration
The Crossbar chip can be integrated on-die with CMOS GPU or CPU circuitry for faster performance.

Cost is of course -- for now -- higher than NAND.  But the greatly improved performance means that it should command a premium when produced.  And Crossbar's ability to produce it on a budget that's a shoestring compared to the giants of the storage industry shows that its technology might not be as much more expensive than NAND as one would think.

V. Intellectual Property, Competitive Concerns Loom

Those are bold words, so is the Crossbar chip -- and RRAM in general -- ready for prime time?  And if so, when?  

The when is a bit easier question to answer. Typically, there's a 1 to 2 year wind up from first wafer-scale prototype tape-outs to mass production, so we may see samples by next year.  And we will likely see a commercial launch in the 2014-2015 window.

Hynix and HP have also promised product sometime around then, but given rocky road that development has taken and the lack of publicly revealed working full chip prototypes, the assumption would certainly be that Crossbar has the lead in time to market.  

The real question is what will happen when Hynix -- and others like Samsung -- do arrive.  The floating gate flash memory industry was largely developed by Toshiba Corp. (TYO:6502) engineers in the 1970s and early 1980s, but it was Intel Corp. (INTC) that first struck commercial gold in 1988.

That brings up a second key question, which also represents a potential problem.  Crossbar must carefully select a fabrication partner.  Samsung seems a natural and wise choice.  However, Crossbar has to be wary that history doesn't repeat itself.  After losing its lead to Intel, Toshiba recruited Samsung to help it mass produce NAND flash.  Samsung would go on to develop its own flash product and leave Toshiba far behind.

Toshiba recruited Samsung for early NAND production, only to be later scooped by its larger fab partner.  Could history repeat itself with Crossbar?

If Crossbar does sign with Samsung -- or Intel, whixh recently expressed openness to serving as a third-party fab -- it must be very worry that its "partner" doesn't turn around and scoop it, leveraging scale and fab expertise to outprice and outcompete its product.

There’s also the question of intellectual property.  HP's Stan Williams says that all resistive RAM devices should be considered "memristors", regardless of the finer technical details of the implementation.  Such a statement could prove a prelude to intellectual property warfare.
Crossbar Technology
Crossbar has over 100 patents pending, but HP has at least five times that number.

Crossbar has filed for over 100 patents on its technology, and has been granted 30 patents.  But that pile is still dwarfed by HP's.  HP reportedly had 500+ patents by 2011 alone, and has surely amassed sure since.  HP and Hynix may have well over an order of magnitude lead in patents over their respective rival.  That could spell trouble for Crossbar should the veterans look to bully it in federal court.

HP has a lead on Crossbar in intellectual property. [Image Source: GUIM UK]

Despite these potential pitfalls, analysts are relatively bullish on Crossbar, given that it's shocked the market with the first mature-looking RRAM product.  Sherry Garber, Founding Partner, Convergent Semiconductors surmises, "RRAM is widely considered the obvious leader in the battle for a next generation memory and Crossbar is the company most advanced, showing a working demo that proves the manufacturability of RRAM. This is a significant development in the industry, as it provides a clear path to commercialization of a new storage technology, capable of changing the future landscape of electronics innovation."
Crossbar product (2)
Crossbar has an early lead on the competition.

Crossbar still has a long fight ahead of it, in a ring of grizzled veteran firms like HP, Samsung, and Intel.  But it's emerged from the first round an unheralded winner.  Not too bad a show for one of Silicon Valley's best-kept secrets, I say.

Source: Crossbar

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

Intellectual Property...
By MozeeToby on 8/7/2013 11:23:06 AM , Rating: 5
HP's Stan Williams says that all resistive RAM devices should be considered "memristors", regardless of the finer technical details of the implementation.
For the love of all that is holy, how do they think they can possibly get away with that? I have a patent on a device that burns fuel to generate work, the finer details aren't important; the fact that my patent describes a steam locomotive and your product is a rocket engine is irrelevant. I hope, but have very low expectations, that the patent judges laugh them out of the court room.

RE: Intellectual Property...
By othercents on 8/7/2013 11:39:02 AM , Rating: 2
That might be true if the idea of using three different materials to create a resistive material for memory usage was already patented by HP. The idea of doing it vs the practical application of what materials is being used are different.

Basically they will have to pay HP for the resistive idea, however HP would have to pay them to use the same material that they use. All other companies would have to pay both to create something similar.

RE: Intellectual Property...
By siliconvideo on 8/7/2013 11:45:09 AM , Rating: 5
The only real winners here will be the lawyers who will make 10s of millions in the up and coming patent battle royal. The losers will be the consumers that end up paying the lawyers through higher product products and delayed technology entry into the market.

Of course Crossbar will make money unless they are sued into oblivion by the big boys. Personally I wish Crossbar does well.

RE: Intellectual Property...
By othercents on 8/7/2013 12:16:35 PM , Rating: 1
The only real winners here will be the lawyers who will make 10s of millions in the up and coming patent battle royal.

Such a pessimistic viewpoint. Companies create agreements over patents everyday without needing to battle it out in court.

RE: Intellectual Property...
By cpeter38 on 8/7/2013 1:22:02 PM , Rating: 4

Take for instance the case of Apple and Samsung ...

RE: Intellectual Property...
By SPOOFE on 8/8/2013 3:44:40 AM , Rating: 2
"Hey guys, let's judge the whole thing based entirely on exceptions to the rule! Nothing can possibly go wrong with that plan! Weeee!!"

RE: Intellectual Property...
By Ramtech on 8/7/2013 1:48:25 PM , Rating: 1
So what?

Patents are used as weapon against competition by corporations against other competition to gain money or damage competition Small corporations will usually wont survive this encounter
Numerous precedents happened use search engine

Take for example MS Android patent licensing This move is making them money and stifling competition to boot. What should be written into stone is that Android would have made without their patents

RE: Intellectual Property...
By SPOOFE on 8/8/2013 3:47:52 AM , Rating: 4
Does someone have a patent on punctuation, and that's why you tried to avoid using it?

RE: Intellectual Property...
By BRB29 on 8/9/2013 8:37:22 AM , Rating: 3
And who writes these agreements, check for legality, and determine royalty payments without going to court?
Last I checked, it was still lawyers.

RE: Intellectual Property...
By Totally on 8/7/2013 7:07:57 PM , Rating: 2
I hate to be that guy but I'd like to believe so but as Apple has proven as long as you have influence and/or someone important in your pocket anything is possible in the legal arena.

By inperfectdarkness on 8/9/2013 1:44:09 AM , Rating: 2
And in other news, HP takes patent advice from Apple's legal team.

if the numbers are correct...
By crispbp04 on 8/7/2013 11:12:35 AM , Rating: 2
this will be disruptive to almost every device market in existence. Phone, tablet, PC, Server, storage... we may see the death of mechanical drives after all! That is, if the comparison chart is accurate

RE: if the numbers are correct...
By Argon18 on 8/7/13, Rating: 0
RE: if the numbers are correct...
By MrBungle123 on 8/7/2013 2:51:33 PM , Rating: 2
You can keep them, I'd rather run an SSD.

RE: if the numbers are correct...
By Flunk on 8/7/2013 4:12:44 PM , Rating: 2
It's not as if anyone is making anyone buy anything here. I bet Hard Disks will be around a good long while.

By Makaveli on 8/7/2013 7:07:42 PM , Rating: 2
lol your the only person still buys raptors!

I don't think anyone wants them from your cold dead hands.

RE: if the numbers are correct...
By tastyratz on 8/7/2013 12:39:27 PM , Rating: 2
I don't know that I believe those charts in that their comparison looks pretty best case vs worst case.
We haven't been restricted to nand numbers like that for a while.
That being said the memristor tech is going to absolutely revolutionize electronics in my opinion. I was excited when this was announced years ago... and I can almost taste it.

RE: if the numbers are correct...
By Etsp on 8/7/2013 4:44:31 PM , Rating: 2
I think those numbers are for an individual chip, not an entire SSD product. It's for the end-product that we usually see the stats for.

By zephyrprime on 8/7/2013 2:13:42 PM , Rating: 2
It seems like it can scale into the third dimension much better than flash. The article implies that flash can be layered easily but this is not the case. Flash can be layered but only by stacking silicon dies on top of each other. This device seems to capable of being layered by just lithography. That really would reduce its cost enough to replace mechanical hard drives.

prototype specs?
By mik123 on 8/7/2013 12:23:49 PM , Rating: 2
Jason, you indicate that the prototype chip they built is 8GB at 25nm.
I looked over Crossbar site, and I can't find that info.

Where did it come from?

RE: prototype specs?
By JasonMick on 8/7/2013 6:43:15 PM , Rating: 2
RE: prototype specs?
By mik123 on 8/21/2013 7:51:49 PM , Rating: 2
I finally found the correct info:

What they built is 1k by 1k array (that's 1MB of memory), built with 110nm process. A far cry from 8GB on 25nm. They say they are going to build something in 2014, which means they have no idea when it's going to be ready.
BTW, that's what HP is currently saying, but again, that does not mean much.

I don't blame you - their marketing was pretty misleading. But, you know, when it's too good to be true...

RE: prototype specs?
By Any14Tee on 8/9/2013 8:02:42 AM , Rating: 2
I know you were asking my namesake Jason (Anonymity in shreds now!) but I have come across the information you were seeking - See below article:

RE: prototype specs?
By mik123 on 8/21/2013 7:55:36 PM , Rating: 2
I found the correct info (see above).

By name99 on 8/7/2013 2:42:07 PM , Rating: 2
commercial-ready ReRAM product

So I can BUY a Crossbar ReRAM product today? No? Two years from now (they hope).
Then it's NOT commercially available, is it?

By tng on 8/7/2013 6:07:27 PM , Rating: 2

It probably is "Commercial Ready" the problem is that they don't have it commercially available yet. Production facilities need to be started, kinks in manufacturing need to taken care of, wholesale distribution, etc...

All kinds of things that need to be done, even if it is ready for primetime.

I will say that this is one of the few things that I see here that is "New and Groundbreaking" that will probably make it to commercial use, unlike the hordes of developments I see on this site by research labs that have no way they will ever make it to a consumer product.

By JasonMick on 8/7/2013 6:50:45 PM , Rating: 3
So I can BUY a Crossbar ReRAM product today? No? Two years from now (they hope).
Then it's NOT commercially available, is it?
No, you are correct, it isn't.

"commercial-ready" =/= "commercially-available"

For example, Broadwell (Intel's 14 nm) chip is commercial-ready, probably since last year. It is taped out. But it is not on sale yet, has not yet entered full mass-production levels, and will not ship to retailers for some time.

Having achieved a commercially ready design with Broadwell, Intel is now just trying to work out the kinks in its production process to scale out full mass production. It will likely do so by next year.

Likewise, this is a taped out chip, which was made on a normal wafer, and by all accounts is ready for a rapid rampump to fabrication once a fab partner is selected.

That rampup will likely take at least a year as I mention in the article, so let's not get overexcited.

At that point the product will be commercially AVAILABLE.

My point is that this product is much closer to commercial AVAILABILTY than anything we've seen. It's already been made reproducibly on a standard process wafer and the design is complete -- all that's left is optimizing production on the fab of choice.

No other RRAM player has demonstrate a chip that close to market readiness. Maybe HP/Hynix have one, but if so they're playing their cards awful close to their chest.

By BRB29 on 8/9/2013 8:40:41 AM , Rating: 2
So I can BUY a Crossbar ReRAM product today? No? Two years from now (they hope). Then it's NOT commercially available, is it?

Commercial ready is not commercial available. Commercial ready just means it is ready for production and the technology to produce it is available. We won't see this RRAM for a while.

read 17 or 140?
By ksenter on 8/7/2013 3:01:12 PM , Rating: 2
In the last image in the article it says read is 17MB/s and write is 140MB/s, in the previous chart it said read was 140MB/s. Which is right?

RE: read 17 or 140?
By JasonMick on 8/7/2013 7:43:02 PM , Rating: 2
In the last image in the article it says read is 17MB/s and write is 140MB/s, in the previous chart it said read was 140MB/s. Which is right?
The 17 Mbps figure is for asynchronous reads vs. standard sequential reads... The write speeds is 140 Mbps, which I believe is also the sequential read speed... But I'll double check that #.

RE: read 17 or 140?
By ksenter on 8/8/2013 9:04:48 AM , Rating: 2
Oh ok thanks. It's just weird how it's displayed in that graphic. Unless I'm missing it I don't see a mention of asynchronous in the article. I haven't looked at the whitepapers though, admittedly. Anyway, thanks for the clarification.

Patent solution is obvious
By Dorkyman on 8/7/2013 2:25:40 PM , Rating: 3
There is a trivial solution to the patent issue. Just make a big contribution to the Dem Party. Presto! No immediate relief, but the secure knowledge that a negative outcome from a patent battle will be overturned by The People Who Know What's Best.

By rs2 on 8/7/2013 7:46:02 PM , Rating: 1
Why are both the NAND chip and the RRAM chip rainbow colored in the comparison image? What color are they, really?

And more importantly, will the interface to these new chips be drop-in compatible with the standard interface used with today's NAND chips? It seems like that would be very important in terms of the viability of RRAM. If device manufacturers have to retool their entire setup to take advantage of RRAM then it will be quite awhile before any affordable, mainstream devices become available.

RE: Rainbow
By tecknurd on 8/12/2013 3:05:56 PM , Rating: 2
IC or integrated circuits in general refract the light because of many layers of substances are used. Also the many different heights on the IC causes more refraction. When the substances are fused to the IC, they create crystals, so that is additional refraction. In this case it is normal for any IC to produce a rainbow effect when its picture is taken or when you see the IC with your own eyes. It is not an artist playing tricks. It is basically the physics of light.

NAND output and input connections are CMOS or about 3.3 volts, so RRAM is compatible with that. They did say that RRAM is easy to manufacture. Finding fab to develop the chip will be a problem because of the many piranhas in the search for the next non-volatile memory.

Hurray for better retention!
By sheh on 8/7/2013 11:11:16 AM , Rating: 2
The rest sounds promising too.

MB or Mb?
By knightmike on 8/7/2013 10:22:47 PM , Rating: 2
One graph shows 140 MB and the other shows 140 Mb. Which is it?

Intellectual property
By jestocost on 8/8/2013 6:54:40 AM , Rating: 2
Where is the problem with the "intellectual property"? It is now turning out that the whole "memristor" story seems to be based on erroneous physical assumptions. Some in the scientific community are beginning to catch on to these problems (see, for example,: “Fundamental Issues and Problems in the Realization of Memristors” by P. Meuffels and R. Soni ( ) and “On the physical properties of memristive, memcapacitive, and meminductive systems” by M. Di Ventra and Y. V. Pershin, Nanotechnology, vol. 24, 2013 ( ).

"What would I do? I'd shut it down and give the money back to the shareholders." -- Michael Dell, after being asked what to do with Apple Computer in 1997

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki