Mannerisms

Intel’s diversification into NAND flash is going to be interesting. It says it’s bringing up its first part, a 4Gbit memory, in three different 300mm fabs and is currently outputting samples. That sounds like a triple-risk, triple-expensive sledgehammer to crack a nut.

Intel says it will add a new fab every year for NAND, which seems excessive when Samsung, Sandisk/Toshiba and STMicroelectronics/SanDisk are all either building or bringing up 300mm fabs for NAND flash, when prices fell over 60 per cent this year and when analysts predict further price erosion next year.

“There’s going to be an almighty collapse in the NAND market”, says Andrew Norwood, senior analyst at Gartner Dataquest.

It will also be interesting in the light of Intel’s previous failures at diversification i.e: two attempts at consumer electronics; two attempts at ASICs, and one attempt each at video-conferencing, programmable logic and chips for mobile phones.

Intel has developed in quite distinct phases. Under Noyce and Moore it was the most admired chip company; under Grove it became the most feared; under Barrett, it diversified; under Otellini, it appears to be faltering, if this year is any measure of what is to come.

In April it forecast a Q2 sales decline - its first forecast of a quarterly decline in five years. In June, it said profits could fall 23 per cent this year and it sold its mobile phone chip division to Marvell for $600m.

In July, Intel made four announcements: a 57 per cent drop in Q2 profits; the lay-off of 1,000 managers; that it might not make its four year forecast; a 21 per cent rise in unsold chips.

In August Intel sold its media and signalling business. In September, 10,500 lay-offs were announced. In October the Q3 results saw revenues grow by 9 per cent, but margins and prices slip. Inventory reached $4.5bn.

Of all these moves, the most depressing for Intel was its withdrawal from the mobile communications market.

It was a major attempt at diversification followed by Intel’s realisation that it couldn’t hack it.

To many people the reason was obvious. Intel’s behaviour in the PC market means that other markets are barred to it.

“The mere fact of Intel’s domination of PC markets is why doors are closed to Intel when it looks elsewhere”, says Jim Tully, vice president at Gartner Dataquest, “people are very nervous about letting Intel extend its reach into other areas.”

“The Wintel experience left people very wary”, agrees Malcolm Penn, CEO of UK analysts Future Horizons, “it’s hard to see how that could ever be allowed to happen again.”

“The whole cellphone industry is paranoid about the Wintel thing,” concurs Stephen Entwistle, vice president at Strategy Analytics.

Meanwhile, heavyweight telecoms chips suppliers like NEC, Qualcomm, Texas Instruments, Freescale, STMicroelectronics, Infineon, Renesas and Agere were strong enough to see off Intel.

This is bad news for Intel. If it can’t diversify, it is stuck in the fast-commoditising PC industry with an increasingly feisty and effective competitor.

In the Wild West days of the semiconductor industry, you’d get fifteen manufacturers investing in enough capacity to each achieve fifteen per cent market share in a particular product area, and think nothing of it.

If, they argued, the resultant over-capacity resulted in driving down prices, then that would work to expand the market. The thinking was that, as prices fell, new applications were found for the chips, so the market grew, so the over-capacity got absorbed, and everyone went away happy.
It hasn’t quite worked like that for sometime as end markets have become saturated, and fewer suppliers has meant there is more control over end prices.

However, with NAND flash, there is, once again, a feeling that this is a product with such infinite promise, such wide potential application and with such a rapid current ability to scale, that it’s worth building apparently massive over-capacity.

After all, non-volatile chip storage is so much more attractive a proposition than disc storage, bringing with it visions of totally solid state, lightweight, instant-on, low power, laptops, tiny camcorders with hours of storage, massive USB dives to store video, computer-like storage capabilities for mobile phones and the like.

These, it is hoped, will be the sort of new applications for NAND flash memory which will be driven by a substantial price fall.

On the capacity side, the build-up is awesome. The Intel/Micron joint venture, IM Flash, is currently bringing up three 300mm fabs for NAND flash at Utah, Virginia and Boise, Idaho. It says it will start building a new fab next year in Singapore and will add others, on a yearly basis, thereafter.

The STMicroelectronics/Hynix joint venture NAND fab in Wuxi, China produces its first wafers this quarter; the new SanDisk/Toshiba fab plans to kick off with 2,500 wafers a month next year; SMIC the Chinese foundry, has started sampling NAND flash, while Samsung remorselessly pursues its $33bn fab-building plan to build eight 300mm fabs between 2005 and 2011.

A wild card in the NAND line-up is Spansion which has announced it is producing a 4-bit-per-cell NAND memory based on the NROM technology it licenses from the Israeli firm Saifun.
Spansion is reluctant to disclose its plans, performance specs, reliability and manufacturing yield for the technology, which it calls Quad-Bit.

“I’m sceptical.”, says Joe Unsworth, principal analyst at Gartner Dataquest, “you have to give Spansion some credit for achieving better output than any of the other licensees of Saifun’s NROM technology. But can they get it into volume production with decent performance and reliability? If they can only do it in 90nm, then who cares, with the rest of the NAND flash industry on 50nm and 60nm?”

Obviously 50nm 2-bits-per-cell is going to be just as dense and just as cheap as 90nm 4bits-per-cell with, very likely, better read and write speeds.

With or without Spansion, the capacity being put on in the NAND business is worrying a lot of people.

“It’s very concerning that pricing is down 66 per cent this year and folks are still adding massive amounts of capacity. We think this could impact both the DRAM and the flash businesses”, says Unsworth.

Prices plummet, but investment in capacity soars, is this madness or do Samsung, Intel, Micron, Hynix, Toshiba, SanDisk and other participants in the NAND flash market know something we don't?

Multi-billion dollar raises in the NAND flash poker game are beginning to cause some of the players to blink.

A final memory of Nuremberg, home to Embedded World last week.

Intel's announcement of a 128Mbit phase change memory could be the company's most significant product launch for almost 20 years.

The news that Spansion and TSMC are getting together to push flash process technology to the limits raises the interesting question of can a foundry compete with an IDM in a mass commodity market?

There were the signs and sounds of a general washing of hands during the long-waited announcement of the ST and Intel flash joint venture.

MoSys, which has licensed its 1T-SRAM technology to the leading foundries, has scaled its technology down to 55nm for NEC Electronics.

The potential for solid state storage in laptops looks pretty enormous with density-boosting technologies like MLC and Quadbit still to be applied to dense, high volume, SSDs.

The underlying fragility of the memory business model has been exposed in all its awfulness by the troubles at Micron.

The reason given for choosing Numonyx as the name for the Intel-ST flash joint venture is that it derives from menomonics, the art of using an acronym, a word, or a phrase to help remember another word, phrase or a list. But could it also be telling us something about the future of non-volatile memory?

After the power outage which saw both memory prices and the share prices of memory manufacturers move sharply upwards, Samsung has, apparently, gone in for some bizarre cost-cutting initiatives.

First it was AMD which licensed the capacitor-less DRAM memory, known as ZRAM, in 2005. AMD wanted to use ZRAM to add memory to microprocessors with less usage of silicon than that taken by a conventional DRAM based on a transistor + capacitor cell structure.

It’s good to see an attempt being made by IBM and TDK to get MRAM into the mainstream by jacking up its density into the multi-Gbit domain, currently enjoyed by tradition floating point flash memory technology.

Spansion’s Mirrorbit technology is the only competitive flash technology which will scale below 45nm, according to the company’s CEO Dr Bernard Cambou.

Surrounded by paddy fields in the middle of Japan’s main island, Honshu, about 170 miles north of Tokyo, is NOR flash memory maker Spansion’s weapon for dominating the NOR flash market, a 300mm fab.

Yesterday's announcement of the purchase of Saifun by Spansion appears to be a good deal for Spansion, because Saifun is priced cheap, at little more than the valuation bestowed by Saifun's current share price, and it gives Spansion total control over who gets access to NROM, the only manufacturable trapped-charge flash technology.

Memory technology still has the capacity to make your jaw drop, and the idea of a chip on which you can store five DVD-resolution movies is truly jaw-dropping.

How Micron and Infineon must be kicking themselves that they didn’t take out Hynix when they had the opportunity.

It’s a funny old world. Someone was chatting to me yesterday saying that Micron was lucky to have its CMOS sensor business, which is a lot more protectable than its NAND flash and DRAM businesses.

Good news for solid state laptops is that Toshiba and Micron aim to move into production with solid state discs (SSDs) next year.

Toshiba is spending its R&D dollars on technologies that will take it beyond floating gate flash, as is shown by the announcement of a trapped charge flash structure which would be suitable for a 10nm semiconductor process.

The deal announced last May to merge the loss-making NOR flash divisions of Intel and STMicroelectronics wasn’t as tightly sewn up as it might have been.

Powerchip Semiconductor of Taiwan, the world’s sixth largest DRAM manufacturer, has joined up with the Belgium-based IMEC DRAM R&D consortium, which is the leading world centre for core DRAM technology development.

Fancy a punt? Rambus shares, now at $16, could soar to $100, or fall to $5, depending on which way a judge decides a court case next month.

The DRAM industry has always been barking mad, but Q4 showed it in one of its spectacularly potty spasms.

It’s surprising that there still are ten DRAM producers after all the consolidation there’s been in this industry sector, but there are, actually, more than ten DRAM manufacturers and, what’s more, the non-top ten had collective revenues of $428 million last year in a market worth $31.5 billion. Here are the ten largest, with their 2007 revenues, as reported by iSuppli:

You have to laugh. First we get Micron CEO Steve Appleton saying that low DRAM prices are a jolly good thing because they’ll drive necessary consolidation in the DRAM industry; then we get the German and EU authorities giving Qimonda a 165 million euro subsidy to build a DRAM fab in Saxony; then we find memory distributors Kingston offering a $40 rebate on a $40 memory card.

Intel’s plan for getting out of its hole in ramping up NAND at a time of plummeting prices is to move to high-density SSDs.

There's an interesting comment on the NOR flash market at http://seekingalpha.com/article/68068-the-spansion-expansion pointing out that Spansion, the only NOR flash player using 300mm wafers, shortly to move to 45nm before its main rivals Samsung and Numopnyx, and with an inherently cheaper technology, is in an exceptionally strong position.

At yesterday’s results meeting, Intel CEO Paul Otellini gave his clearest hint yet that he’s considering dumping Intel’s stake in the IM Flash joint venture with Micron Technology.

If ever there was a miserable story in the history of the semiconductor industry, it is the story of Rambus.

On the face of it, it sounds utterly daft to use NOR flash for data storage. The cell is too large and the read speed too slow. But this is exactly what Spansion is proposing.

It was sad to hear about the death of J.R.Simplot, who put up the founding money for Micron Technology. He had a good innings, dying in May at 99 after the most colourful life.

How on earth do you get the price of commodity chips to go up? Answer: Create a fear of shortage.

Would you Adam n' Eve it there's a debate going on about whether SSDs use more power than HDDs?

If, and it's a Big If, SanDisk has cracked the problem of  how to make re-writeable 3-D flash memory, then it has a dilemma: should it license the technology to Samsung?

"We are working with utmost urgency on the question how to close the chapter on Qimonda as quickly as possible," Peter Bauer, CEO of Infineon, tells the German newspaper Sueddeutsche Zeitung.

Rambus is looking to cut jobs, monetise flash interfaces, sue Nvidia for treble damages and get a legal order stopping Hynix selling DRAM in the US.

Phase change memory has to move from a transistor array to a diode array, has to add non-volatility to a diode array, and will have to master multi-bit per cell technology before it can be a commercial product, according to the CEO of Spansion, Bertrand Cambou.

Trapped charge technology can extend down to 18nm, according to Spansion, which believes it will have a lock on the technological future of flash memory when traditional floating gate techniques run out.

It's not how the size but what you do with it that counts, and Spansion is going down a counter-intuitive application route for what it hopes will be ultra-small die, ultra-high bit-per-cell, ultra-cheap memory.

As it further understands the potential of its trapped charge Mirrorbit technology, Spansion is extending it to new applications, such as data storage, DRAM replacement in servers, DRAM replacement in  wireless handsets, and ultra-cheap content delivery.

With SanDisk's share price well below the offer price for the company by Samsung, it seems that the markets do not hold out a lot of hope for the takeover attermpt being successful.

Interesting to see that NXP is working on phase change memory with a little UK start-up called Ilika. Memory is the elixir of life in the semiconductor industry - when it's not the harbinger of death.

Two of the IC industry's cherished rules appear to be being broken by the parlous state of the NAND market: Rule One: While revenues may decline the growth in unit volume rarely does; Rule Two: People always want more storage capacity.

With the awful examples of Rambus and Qualcomm in front of it, it's a bold move by Spansion to decide to sue Samsung for patent infringement in a bid to establish itself as an IP licensing company.

Has IM Flash Technologies sneaked a process technology lead on its rivals in the flash market? It certainly looks that way.

Toshiba blinked first. Reports in Japan that Toshiba is considering postponing two new NAND Megafabs are just a face-saving way of getting the news out. It's also a capex-saving way. By not starting the fab-building next year, Toshiba will save a good chunk of its $3.9 billion budgeted capex for the financial year.

Qimonda could be starting a whole new model for the DRAM industry. Apparently it has asked the German government for financial help under its 500 billion Euro bail-out plan for German industry, while the State of Saxony may take a direct shareholding in Qimonda.

Why do they do it? Well about twice a decade the DRAM market comes good. Demand outstrips supply, prices are up, the money comes rolling in, you're on a high like a heroin fix or a win on the horses. You never forget it.

Hynix could be showing the way for industry in the expected downturn. It is planning to sack a third of its executives.

Like dogs, every now and again the DRAM industry has its day. It's always been like this. The manufacturers make massive losses for years, the market remains over-supplied, the prices slump below production cost and then, like rain in the Sahara, everything blossoms at once: prices stop falling; demand soars, supply is limited and everyone makes an absolute packet.

Could rain be about to fall in the Sahara?

Is phase change really going to happen? For a while now people have been looking at phase change like they look at ferroelectric memory, bubble memory and MRAM as a technology which was never really going to make it except in niches. But yesterday things changed.

Well is phase change going to happen or not? After quoting Numonyx's CTO last week that Numonyx started selling a phase change chip last year, I have had a number of people writing in to say this doesn't actually seem to be the case.

Giving a hostage to fortune is not usually a very good idea, especially if you don't have to do it.

The only people with a greater insouciance toward the value of money than Wall Street bankers are DRAM manufacturers. To them money is as precious as confetti.

The interesting thing about the alleged flaw in Intel's solid state disc drives is not how they fix the flaw, but how they react to it.

The change of emphasis at Toshiba - from semiconductors to power generation - has coincided with a change at the top. The new President will be an executive from the heavy electrical machinery division who is an expert in  nuclear energy generation, corporate senior vice president Norio Sasaki.

Freescale has kicked off the jolliest wheeze for years with semiconductor companies buying back their debt for a fraction of its par value. The latest to do this is the Taiwanese DRAM company ProMOS Technologies which has paid off its convertible bondholders for 25 cents on the $.

In February I did a post musing on whether phase-change memory is ever going to happen, or if it is a sort of Techno-Ponzi scheme for boosting your share price, persuading your management into investing R&D funds, or for wooing VCs.

Great to see a new memory technology emerge. This is the sort of stuff which used to make the semiconductor industry constantly exciting and, true to the tradition of new US semiconductor technologies, the new CMOx memory technology, pioneered by Unity Semiconductor, has come from an American venture capital-backed start-up company.

A Terabit memory chip in six years?  That's ahead of the ITRS, ahead of Moore's Law, and ahead of people's expectations for non-volatile memory which is not expected to scale much beyond 32nm

A memory technology to beat all memory technologies has been developed by the Lawrence Berkeley National Laboratory and the University of California, Berkeley.