Gathered below are the top five stories for the week beginning Sunday April 25, 2004, as ranked by Silicon Strategies' readers, up to and including Saturday May 1. The ranking is based on the number of reader "views" or "hits" on a particular article.
Mark LaPedus, the editor of Silicon Strategies, and Peter Clarke, news director, provide commentary and analysis on the week's top stories.
1) TSMC's booming Q1 profit revealed after mistake .... After parts of the first quarter financial results of foundry Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) were disclosed to the Taiwan Stock Exchange ahead of schdule by mistake, the company published its full Q1 results without commentary, revealing booming profits on sequentially flat sales.
Athough profits are booming at TMSC and the first calendar quarter in the foundry usually weighs in with a seasonal lull, those flat sales were still a slight cause for concern. If ASPs went up that must mean that wafer shipments fell, or vice versa. We had to wait to Friday to find out more. PDC
2) Tech analysis: TSMC reworks low-k process .... On the way to adopting low-k dielectrics, the world's largest foundry has run into a problem, collapsing vias. EETimes' Dave Lammers dissected a low-k failure on TSMC's 130-nanometer manufacturing process -- a process node for which many people thought the low-k problems were behind us.
Lammers quotes TSMC executives arguing that the problem is isolated and occurs on very few chips, but he points out that moves to 90-nm -- where low-k is expected to be the norm and FSG the exception -- are already taking place and the work in hand is meant to be finding solutions to 65-nm problems, not problems at 130-nm.
Also the solution had a familiar ring to me. TSMC went back to its library and memory compiler vendors and asked them to create redundant vias in SRAM cells. Apparently adding metal strengthens the overall structure and prevents thermally-triggered via collapse.
But wasn't it the case that some years ago it was reported that chemical mechanical polishing was causing "dishing" of large low-k dielectric areas so that the CMP was no longer producing flat surfaces ready to for the next layer of interconnect. The solution, as I remember, was to add unconnected vias to produce a more uniform surface across the chip. The subsequent problem was that the additional metal added to the interconnect capacitance thereby reducing the benefit from using a low-k dielectric, and was not taken into account during timing-closure calculations, allowing chips that were now manufacturable to start failing in that regard.
Surely, at the very least, adding redundant vias to SRAM cells, whether connected or not, is going to negatively effect circuit performance? PDC
3) Heard on the Beat: Hynix dumps DRAMs, IC prices fall .... Fusion Inc., an independent distributor, tipped us off to a significant drop in double-data-rate (DDR) SDRAM pricing which was also the subject of our weekly DRAM Bulletin from iSuppli Corp.
For weeks -- if not years -- DRAM makers have promised that shortages are coming. Well, the DRAM shortages are just around the corner (and the check is in the mail.)
Prices for 256-Mbit parts have fallen amid concerns about a slump in demand. The possible cause? There are reports that Hynix dumped a slew of 256-Mbit DRAMs in China. Not too surprising if true. China has always been a dumping ground for excess inventory. But what's surprising is the current supply/demand situation, leaving some to believe there are still too many memory makers serving too few customers. ML
4) Tech analysis: conflicting signals at 90-nm .... When Agilent Technologies Inc.'s ASIC products division first moved chip design from 130-nanometer to 90-nm, signal integrity got worse by an order of magnitude. EETimes' Richard Goering looks at one of the issues driving 90-nm chip failures -- and discusses potential solutions.
At one time, many believed that it would be a slam-dunk in terms of migrating chip production from 130- to 90-nm. After all, the chip industry learned plenty of hard lessons at 130-nm, especially materials integration.
However, 90-nm is turning out to be harder than previously expected. "Signal integrity was really an order of magnitude worse," said Jay McDougal, microprocessor design methodology manager at Agilent.
McDougal's experience seems to track with that of other users and matches what EDA vendor representatives are saying. Such problems as crosstalk-induced delays, crosstalk-induced glitches and power noise due to voltage drop are all accelerated at 90-nm, making design closure more difficult.
It leaves one to believe that 90-nm designs won't become mainstream for a long, long time. Intel, TI, and a few others are on the leading-edge of chip design, but Moore's Law is fast becoming the exception and not the rule. ML
5) Analysis: is Infineon pulling back from outsourcing? .... As Infineon laid out plans to increase investment in manufacturing Peter Clarke asked the question and reminded readers of the background. One reader chipped in with additional information here.
Monday, April 26
ASMI to acquire Korean ALD equipment vendor .... Making its second acquisition in recent weeks, ASM International N.V. Monday said it has acquired Genitech Inc., a South Korean supplier of plasma enhanced, atomic layer deposition (ALD) equipment.
ASMI has struck again, this time, acquiring a South Korean ALD tool supplier. The Dutch-based equipment vendor recently bought NuTool, a supplier of electroplating tools for copper integration. The two acquisitions enable ASMI to provide the PVD/ECD technologies for next-generation, copper/barrier seed applications.
Good moves by Arthur del Prado, president and CEO of Dutch-based ASMI. At the same time, ASMI is making a run at becoming the next Applied Materials, that is, being a one-stop shop of chip tools for future CVD/ECD/PVD applications.
Rival Novellus is making a similar bid. Will Novellus and ASMI become the next Applied Materials or simply the poor man's Applied Materials? Time will tell.
Meanwhile, ASMI continues to loss money amid a restructuring and cost-cutting strategy. In 2003, front-end equipment represented 49.2 percent of its total net sales. Assembly and packaging equipment represented 50.8 percent of consolidated ASMI net sales.
One of the problems is ASMI's front-end business is basically tied to one customer--Intel. Intel uses ASMI's low-k tool and material for its 90-nm process. Intel reportedly has also qualified Applied's Black Diamond film for 90-nm, but the chip giant has only put ASMI's technology into production.
Intel is also using ASMI's epi reactors for strained silicon production. On the boring backend side, ASMI has overtaken K&S in the wire-bonder space, but K&S claims it has regained share.
By the way, rumors were circulating that ASMI was looking to buy an ALD equipment vendor that started with the letter "G." There was wild gossip that surrounded a blockbuster merger between ASMI and Genus. Turns out that ASMI bought another firm that started with "G." It bought Korea's Genitech for a relatively small price, as compared to possibly the hefty the price tag for Genus. ML
SMIC scores 29 percent growth, profit in first results .... Chinese foundry Semiconductor Manufacturing International Corp. (SMIC) made a net income of $8.6 million on sales revenues of $186.9 million in the first quarter of 2004, the company announced. Sales increased nearly 30 percent from the prior quarter.
Tuesday, April 27
Foundry production on the rise for analog market .... Foundry production for analog devices is on the rise, according to a new report issued by Databeans Inc.
Wednesday, April 28
VC start-up raises $250 million to spend on spin-outs .... Garnett & Helfrich Capital, founded late in 2003, has launched its Venture Buyout (VBO) fund, with initial funding of $250 million raised from universities and institutional investors. The fund is intended to finance the 'spin-out' of under-performing and non-strategic businesses, product lines, and divisions from public companies in the fields of communications, Internet-based media, semiconductors and software.
Nice work if you can get it. I would just like to draw attention to an interesting disparity here.
An engineering start-up is made to jump through hoops and disclose a detailed business plan to get dribs and drabs of $1 million, $2 million or $10 million from VC companies. But a VC start-up that says it is going to go and find businesses that aren't doing very well and invest money in them, gets given a quarter of a billion dollars to play with.
Indeed, there may well be many more VC companies looking to invest in technology than there are technology start-ups. Could that drive a need to prise divisions, and product groups away from their parents so that they can be taken private and then be IPO-ed or sold-off all over again? Ker-ching. PDC
Thursday, April 29
Acting CEO denies Trikon is in a "death-spiral" John Macneil, the acting chief executive officer of semiconductor manufacturing equipment vendor Trikon Technologies Inc., denied the company is in a "death-spiral" during a conference call on Tuesday (April 27). He said company business models showed Trikon would hit break-even by the end of 2004 or early in 2005.
The dog that did not bark during the conference call on Trikon's most recent results was whether there were any predators or white knights in the wings preparing to make an offer for the company.
We've started to see merger and acquisition activity increase in the semiconductor equipment market and Trikon, despite protestations about a happy customer base, is looking like an acquisition target. The customers may be happy but not enough of them are putting their hands in their pockets.
If the Orion ultra low-k technology is as good as has been claimed, then the tide may indeed turn in Trikon's favor one day. But it takes more than world-beating technology to beat the likes of Applied Materials. One day soon a TSMC, an STMicroelectronics, a Freescale, or an AMD -- or all of them together -- may go to one of the large equipment vendors and say: "We would like you to be our low-k dielectric technology supplier at the 65-nm manufacturing process node but we don't want your technology; We want you to supply Orion. Make it so." PDC
Friday, April 30
TSMC sees consumer, communications driving strong Q2 ramp .... After reporting booming profits in the first quarter of 2004 earlier in the week Taiwan Semiconductor Manufacturing Co. Ltd., said it expects the second quarter to show wafer shipments increasing by close to 10 percent sequentially and average selling prices to improve by a low single digit percentage sequentially.
TSMC keeps humming along despite a hiccup in ASPs for Q1. Overall, TSMC is expected to set new records for revenue and net income in 2004, the company said.
Capacity utilization could exceed 105 percent in the second quarter. How is that possible? The steppers, etchers, and other fab tools must be working overtime in Hsinchu. TSMC is the foundry that never sleeps. ML