Samsung Semiconductor Europe has delivered a moving eulogy for the sadly departed Japanese DRAM manufacturer, Elpida. But, just because it shed a tear or two doesn’t mean it wants to buy out any of the bankrupted company’s shares.
Speaking with TechEye, Gerd Schauss, director, marketing intelligence said it’s never nice for him to see a company with great technology pass away, but that is the situation. “The economies of scale doesn’t help them,” Schauss said, “the production location in Japan is certainly not a low cost area, and it’s pretty hard to be competitive.” Different days, Schauss points out, from the 90s when Japanese companies represented 70 percent of the DRAM market. Elpida was the remaining manufacturer.
Sad though it is, Samsung declared that it is in no way interested in taking over Elpida or becoming a major shareholder. “Samsung has grown organically over the past 30 years and from that perspective we are not pretending to change that,” Schauss said, “the growth in the next year – business will mostly come out of organic growth.”
In an industry which some analysts say has been on the ropes, Samsung Semi hasn’t found itself struggling. Although industry commentators have been talking about a short to long term domino effect since Elpida filed for bankruptcy – and depending on the capacity it runs at – Schauss doesn’t expect any impact on production. “In the mid range,” though, “you never know because when you solder down DRAMs you are taking what is qualified in a specific platform, and if a manufacturer has only qualified Elpida and another for a specific product, there might be some short term impact”.
Schauss explains that now, more than ever, it is important to make products which are energy efficient and as green as possible. Over the past years, for a server running a gigabyte of system memory, energy consumption didn’t factor in. As virtualisation has become more prevalent, though, more memory has to be put into a single server. It’s a clear cut case in the enterprise. SAP, says Schauss, will have as a minimum 512 gigabytes of system memory, “then, all of a sudden, it plays a very, very important role.” Even in deployments with 96 gigabytes, you have a situation where memory represents about 25 percent of the total system, Schauss said.
Working with partners in the industry, like manufacturers and software companies such as SAP and Oracle, Samsung thinks lower processes and higher efficiencies are translating into a benefit that the end user which is easy to understand and the benefits are crystal clear.
Schauss played down the importance of differentiating between processes at a certain point. “How often can you split a hair before you come to that process size on a chip? If you save megawatts of energy, I think that’s something everybody can understand.” Samsung Semi now calls its DRAM products 20x – as in they’re twenty-somethings, although not the same sort you’d see in Starbucks.
The closest competition is dwarfed by Samsung Semi, according to Schauss. He reckons that Samsung products are at the least, one and a half generations ahead of others in the market. The reason for this is, simply, Samsung has grown organically into an enormous company which has the resources to invest in its own research, using its own manufacturing. Samsung is able to throw cash at innovating on its process technology – typically $6 to $10 billion. So, says Schauss, “when another manufacturer is able to invest $1 or $2 billion, he can only partially upgrade his equipment”. For example, when a customer buys into Samsung’s 35nm technology, Schauss claims they are buying from brand new lines because the company is able to run full, thorough upgrades. At other companies, Line A could be at 40nm, line B could be running at its peak at 30nm, but the average process being ramped could be at 48nm. “In our case, we have already kicked out all the old equipment and are using it for something else – with DDR3, it is at 35nm”.
Schauss says that market conditions mean the historical top user of DRAM, the PC, will see single digit growth. It “will not be double digit growth again”. A reason for this is because OSes don’t require more than 4GB. Even if some system makers are putting in 6GB or 8GB for differentiation reasons, he says, in most cases it’s not necessary unless you’re a heavy gamer or have specific applications. The tablet and smartphone boom does not offer itself as a growth engine for DRAM, either.
However, in servers, DRAM will see “massive” growth. This will lead to DRAM bit growth over the next couple of years below the long term average. It will also lead to pressure on the manufacturers – when you introduce a new die string, you introduce a better performing device, Schauss says, but you also produce more devices, so you automatically bring in yet another bit growth on the production side, which makes operations more complex. “If you want to play in the 1st league,” he says, “you have to invest and produce more”.
As for differentiating, the next obvious step for computing is DDR4 and low voltage devices. Samsung says it is striving to keep the quality high, specifically in the server environment, where there are so many components propping up even 100 GB of memory. And if just one component fails, the technician has to come out and find the problem. With customers like IBM and HP, they can’t afford to let that happen and need the best quality. This is where Samsung is winning business in an otherwise interesting marketplace for DRAM.
Overall for the industry, Samsung remains confident. “At the moment,” Schauss says, “there is something like a minimum of 12 weeks of inventory. A whole quarter is huge. So normally, when you have one and a half, two percent oversupply, prices fall, like a rock. With 12 weeks plus, we’re not expecting any short term change.”
“In my opinion, the very earliest time where a change could happen, would be the fourth quarter when production ramps. Though I’m not really expecting anything there, the good thing with shortages is they always come as a surprise,” Schauss said.