Thursday, July 21, 2011

When Getting Bigger Makes You Smaller

I've been spending some time with the presentations given at Semicon West this afternoon. What I am particularly interested in is the amount of money companies are going to have to shell out to stay in the game. Years ago I wrote some pieces about the memory business - calling it a "No-Limit-Poker" game. This particular poker game had a twist - the ante went up every hand and you had to play every hand. If you left the table, you were out of the game.

Well, that analogy held true over the years. In the mid-to-late 90's the rising stakes flushed the Japanese right out of the DRAM business. Sure, Toshiba is in the memory game but at that time I was focused on DRAM. NAND Flash was just a blip on the radar screen.

Does the analogy hold true today?

After Intel said they were raising their capital spending budget (again) I thought a bit of a review of some of the predictions from last week would be appropriate. Yesterday I pointed to a 450mm panel discussion at the MCA BrightSpots Forum. Today, I've dredged up some comments from the folks at Gartner.

At the 450mm Panel Session Bob Johnson (no relation) had this to say:

Why 450mm?

  • Equipment for each node costs more than previous nodes

  • Periodic wafer size increases needed to compensate for more expensive nodes

  • Can save 30% on production costs (target)

450mm has nothing to do with Moore’s Law. It's all about economics.

Later in his presentation he mentions the narrowing number of players:

Less than 10 companies are currently building leading edge fabs:

  • 1-2 Logic IDMS (Intel and Samsung)

  • 2-3 Foundries (TSMC, Globalfoundries, plus UMC or SMIC)

  • 4-5 Memory Companies (Samsung, Hynix, Toshiba (w. Sandisk), Micron/Nanya/Inotera, Elpida/Rexchip

The semiconductor equipment companies say they will be ready when customers decide to make the move. I am sure they will - after a lot of kicking and screaming. The development costs are going to be enormous. Several suppliers that I talk with on a regular basis are now saying that 450mm has been moved up on their priority list.

At another session called "Contemporary Packaging: Challenges and Solutions for 40nm and Beyond", Gartner's Jim Walker served up some more food for thought:

The Bottom Line on Design Complexity is:

  • Increased costs combined with reduced growth will cause significant consolidation in chip vendors…

  • The industry will lose nearly one third of all semiconductor vendors by 2020

Wow! One-third going out of business? Well, that must all be due to, yes, you got it:

The Increasing Scale of Manufacturing

Here is the bottom line:

  • Between the 45nm and 8nm nodes, logic fab costs will double to $10 billion. But a 31x decrease in cost per transistor will be maintained.

  • Only four companies will be able to follow Moore’s law by 2018

  • The annual number of new fabs built will fall by 60% between 2011 and 2015

  • By 2015 foundries will account for about one third of the value of all semiconductors compared with about one quarter today

  • By 2012, over 50% of packaging/test (SATS) will be outsourced

  • By 2015 collaborative R&D will save the semiconductor industry over $30 billion in annual R&D expenditure

Well, ain't that just peachy? We'll be down to four companies that can pursue Moore's Law by '18? So, all the development work for advanced lithography (EUV) - in addition to the 450mm transition - should really come out of 4 buckets. I mean, who else other than Intel, Samsung, TSMC and perhaps Global Foundries, has the skin to step up and play that game?

I can see the number of new fabs falling off - I have no issue with that. Why build a new fab when you can buy an old one? Old fabs and fab tools don't die - they just get refurbished and reused. It doesn't just happen with really old stuff, it even happens with 300mm wafer process equipment. Texas Instruments is fine example - they are cracking out analog parts with tools from the first 300mm DRAM fab (the fab was once owned by Infineon and then eventually spun out as Qimonda - TI bought the equipment for a fraction of what it originally cost in a firesale).

I have a lot of questions about these predictions and I am definitely open to hearing what others think.

Feel free to chime in!

No comments:

Post a Comment