iXBT Labs - Computer Hardware in Detail






November 2001 Hardware Digest

Two adjacent still strikingly different months - October and November. If the first is the time of new technological announcements and industry development directions, the second rather has an "application" shade. Well, technologies appear as always, companies announce new boundaries for one/two/five years; nevertheless, all this takes place on the "background". And the forefront is occupied with the most beneficial period of the year, pre-christmas sales. The main events of the month, Microprocessor Forum in October and Comdex Fall 2001 in November, give the exact image of events.

But let's be consecutive.


The Month began with a statement from Intel's CEO Craig Barrett that the company was going to speed up the development of new technologies for processor manufacture and to start using 90 nm (0.09 micron) manufacture right at the beginning of 2003 (!). If you remember, earlier Intel mentioned 2003 as only a year to launch 0.09 micron technical process without any specific terms.

As you know, new Intel's 90 nm technical process is has two names: P862 and P1262. The first is intended for work with 200 mm plates (pay attention to the first digit of the process name, it means diameter in inches), P1262 with 300 mm accordingly. Certainly, they will not do without the latest features of this area - copper connections, low-k dielectrics, technologies for creating the most precision transistor gates.

Intel has things to be proud of here: in the middle of the month it announced the creation of the smallest transistor gate of 15 nm width, and closer to the month's end announced a new direction codenamed "TeraHertz".

Transistors with this structure will appear in Intel's chips with clock frequencies from 10 to 20 GHz by around 2010. The structure of TeraHertz transistors assumes use of new zirconium oxide dielectrics, plates with epitaxial layer and silicon-on-insulator (SOI) technology. First processors with such transistors, according to preliminary estimates, are expected in 2005. Nevertheless, some elements of TeraHertz technology will be launched before 2005 along with launch of 0.09 micron manufacture.

It's not a secret that new developments are primarily directed at reducing heat-evolution of future processors to keep it within 100W. According to preliminary information, TeraHertz technology will become basis for P1268 technical process.

Year comes to end, and many analysts guess at Intel's next-year capital expenditures. To maintain rather intense schedule of launching six new 300 mm factories, it will have to spend at least $7 billions in 2002. As many think, after wearisome and not so profitable 2001 such sum will be rather costly for Intel; nevertheless, the company will most likely go for these expenses (about 18% of Intel's capital), as it won't allow itself to lose the leadership, and according to history of economy, recessions are temporary.

I think company's capital expenditures will strongly depend on tendencies of restoration of entire semiconductors market, as Intel is present at almost all sectors of semiconductor industry.

General forecasts for industry development in 2002 are gradually becoming optimistic: Gartner Dataquest, for example, considers that in 2002 chip manufacture and shipments will become stable and even expects increase of sales volumes. Still by just 3% in comparison with 2001 from $147 billions to $152 billions. Year 2003 forecasts are more optimistic gross sales volume is expected to raise by 30%. 2002 won't, certainly, become a "rescue" year for high-tech companies, it will rather be a continuation of strength tests. Obviously those will survive who despite numerous difficulties, saving on everything and laying off staff, will invest in capital construction, expecting the future raise.

Such situation is hardest for lithographic equipment manufacturers. Development of unique tools for chip manufacture is always costly, and sales volumes of such equipment, shipped apiece, decrease month by month. Japanese Nikon, for example, has already announced that in 2001 it will sell 10% less equipment than it has been planned before. Certainly, it won't do without harsh measures: by March 2002 (Nikon's fiscal year end) about 50% of 3100 temporarily contracted workers will be laid off. Such industry leaders, as ASM Lithography, Canon and Ultratech Stepper, have already informed about reduction of expected incomes.

Intel, planning to launch the whole constellation of new factories in 2002, could have become a "magic wand" for Nikon in terms of selling new 193 nm scanners. However, ASML as well as Canon hopes to receive a fair share of these orders. Most likely, good luck will accompany the one to offer more perfect equipment and more rational delivery terms.

SVG, that is now a division of ASM Lithography, but with its own line of lithographic equipment, business is bad as never (see our summer editorial). Just about a week ago ASML's representatives announced that development and release of 193 nm SVG Micrascan V scanners is cancelled (such equipment will be released only by head company, i.e. ASML), and SVG will turn all efforts to development of 157 nm scanners. And here's the next trouble: release of 157 nm Micrascan VII scanners, that SVG promised Intel to ship in the beginning of summer, are officially postponed again - to the end of 2002. The reason is the same: problems with CaF2 lenses.

Let's remind, that due to delays with deliveries of new scanners Intel already had to correct its plans, in particular, to begin to release new 0.13 micron processors on old DUV equipment, but with application of expensive enough phase-shifting photomasks. Probably, now Intel will really consider offers of 193 nm scanners from ASML, Canon, and Nikon, and new 157 nm scanners from SVG will be required for 300 mm plates with use of 65 nm norms (with 35 nm gates) for P1264 technology, expected to be launched closer to 2005.

The second place in technological news rating I'd like to give to Taiwan TSMC: in the near future it is going to add five new fabs to Fab12 and Fab14, already mass-producing 300 mm silicon plates. All five factories will be located in Taiwan, construction costs will make about $13 billion. For the nearest ten years the company's capital expenditures are planned to about $18,5 billion. TSMC plans to invest some of them into one of China fabs.

Now TSMC does not have problems with manufacture load. Its 0.18 micron capacities are filled up with orders. And, according to company management, the third quarter was completed profitably - rather good for 2001.

Thus, news of month's end that NVIDIA is going to transfer a part of orders to TSMC's main competitor - Taiwan UMC were like a bolt from the blue. A part of NVIDIA's 0.15 micron production (chips for XBox) will be manufactured by UMC, another line of 0.15 micron chips will be still produced by TSMC capacities under the old contract. However, probably placing of additional orders at UMC enterprises is only one of NVIDIA's measures for additional stability, and not disagreements with TSMC, rumors about which were being exaggerated by some journalists during the second half of November.

As if to future orders for 0.13 micron manufacture, they are still unclear: 0.13 micron manufacture of both UMC, and TSMC is generally pilot. Actually TSMC already produces consignments of 0.13 micron Ezra processors for VIA and has announced its full readiness to launch such manufacture of Crusoe TM5800 chips (as we know, the question is about the final version of Transmeta's technical documentation). However, yield of such processors is not comparable to scale of NVIDIA's orders. Moreover, those to get orders for mass production of new 0.15 micron NV17M, NV17, and NV25 graphic chips are undecided as well.

Having spoken about UMC's business, let's proceed to a "surprise of the month": in the beginning of November AMD has announced at the meeting with financial analysts that it is going to use third party capacities for processor manufacture since 2003 to compete with Intel better. It's primarily interested in those chip manufacturers able to provide 90 nm (0.09 micron) processor manufacture by the beginning of 2003. as by this year Dresden Fab30's resources are expected to be completely exhausted.

Certainly, AMD does not terminate further fab construction. By 2005 it plans to launch two new 300 mm silicon plate enterprises. One of them is at present being constructed jointly with Fujitsu - Fab JV3 (Japan) already mass-producing flash memory, the second is being constructed together with still unknown partner and will be engaged in microprocessors manufacture. AMD's management affirms they already have pilot samples of Athlon processors made "on the side", and performs qualification tests now, refusing to name the partner. However, according to analysts, there's a great probability this partner is UMC.

Another AMD's unexpected move: in November it signed a contract with Soitek, SOI silicon plate manufacturer. Unlike competitors, Soitek has established mass production of chips on 200 mm plates with SOI, and now the company owns about 80% of this market.

Meanwhile, the question is the release of plates for AMD's Hammer CPU manufacture. The latter manufacture will, certainly, be deployed in Dresden where, according to the latest information, AMD and Infineon may create another joint venture to reduce R&D expenses. There's no exact data, but new enterprise is expected to include a factory and a research centre.

Tokyo Electron and Toshiba this month reported about progress in development of < 0.10 micron technical processes: together they achieved smaller coefficients in dielectrics for use in chip manufacture. The new method of plating of low-k dielectrics onto surface of copper connections allows to perform this with a 19 electron beam tubes that allows to lower copper solidifying time to 2 minutes at 35C temperature - much faster than 30 minutes in the special furnace at 400C when it is impossible to get rid of partial copper-into-dielectric diffusion. As a result, these manipulations allow to achieve stable k=2.91 suitable for 0.10 micron process.

Fujitsu announced annealing process improvement in laboratory conditions, based on deuterium application that should solve problem of "hot charge carriers" stability in the future chips. Fujitsu already uses such 0.13 micron process in its laboratory, though actually it's optional as other companies think. Nevertheless, tests have shown that such technological process allows to increase microcircuit life almost twice in comparison with usual hot hydrogen processing. Actually, deuterium technology is already applicable to 0.09 micron chips.

In the beginning of the month Motorola announced the creation of new ThoughtBeam division that will engage in development of technology for creating a superfast computer chip. This technology will allow to combine silicon with gallium arsenide (GaAs). In spite of the fact that silicon is unpretentious and cheap, it greatly slows down optical systems. Motorola is not a beginner In this area as it already has a load of patents and now prepares to produce enough SiGaAs plates for potential partners to estimate advantages of this technology.

HyperTransport Technology Consortium lead by AMD has considerably increased in November. Texas Instruments, that is going to use HyperTransport in a new generation DSP chips, was the first to join. Comdex Fall 2001 brought another 14 participants: Acer Laboratories Inc (ALi), Altera, AMCC, Fast-Chip, Flow Engines, GDA Technologies, Josipa Company, LEDA Systems, Marvell Semiconductor, Nokia, Spinnaker Networks, Teradyne, Xilinx, and 0-In Design Automation. You must admit that together with such leaders as, for example, API NetWorks, Apple Computer, Cisco Systems, PMC-Sierra, NVIDIA, Sun Microsystems, Transmeta, and Silicon Graphics this list commands respect and confidence that HyperTransport technology, that we can already estimate, for example, in NVIDIA nForce chipset, will have a long life.

There was some sad news as well. Matsushita decided to close its mobile devices fab in Tratham, Great Britain and to consolidate forces in Czech Republic where a new more profitable factory will be opened in the near future. NEC has announced the decision to outsource two Japanese telecommunication equipment manufactures, NEC Miyagi and NEC Yamanashi, to a still unknown EMS vendor. The company plans to finish this by the year-end. The reasons are known and old: reduction of sales volume and profitability due to economic downturn.


The majority of processor announcements took place at the beginning of the month and the eve of Comdex Fall 2001. Intel announced release of Pentium III Tualatin processors for the market of thin servers and notebooks; S and M models of LV (Low-Voltage) 0.13 micron Pentium III 700 MHz with 512 KBytes cache; and other processors of this family.

LV Pentium III has FSB 100 MHz, working voltage is 1.1V. S modification of such processors is intended for small servers where blades are located very closely to each other and the element density is very high. Pity, these novelties still do not support SMP mode.

As to the further development of this line, 800 MHz LV and ULV versions are expected in February 2002, and then - 900 MHz ULV and LV 933-S processors, all of them supporting dual-processor mode. M version for notebooks will be released about the same terms.

1.6 GHz Mobile Pentium 4 expected in mass amounts closer by February next year in 1.4-1.7 GHz versions were shown on Comdex Fall 2001 for the first time. It was also announced about the reduction of mobile Pentium III and Celeron line. Release of and orders for 0.18 micron 800 MHz Mobile Pentium III and 700 MHz Mobile Celeron were henceforth cancelled.

Intel delayed as well in the sector of "heavy" 64-bit server processors: Compaq has suspended manufacture of Proliant PL590 systems on 733 MHz and 800 MHz Itanium, motivating this by problems with standard tests passage. McKinley line is expected without such problems.

As a whole, Intel keeps promises and preliminary roadmap: main announcements of new processors are expected new year.

AMD announced several novelties as well, like FSB 266 MHz Athlon XP 1900+ processor (see "AMD Athlon XP 1900+ (1600 MHz) processor" article), traditionally manufactured on 0.18 micron technical process on Fab 30, Dresden.

Another announcement: a novelty of Duron line on Morgan core - 1.2 GHz version, unmarked by PR-rating unlike Athlon processors. New 1.2 GHz Duron, as well as its predecessors, is equipped with 192 KBytes cache (64/128 KBytes L1/L2 cache), 3DNow! Professional technology. This novelty will be also manufactured at Fab 25, Austin.

At Comdex Fall 2001 the company has for the first time shown a prototype of Athlon XP processor with Thoroughbred core, expected to appear already in 0.13 micron variant with clock rates (er... rating) from 2200+, its mass deliveries are to begin in the first quarter of 2002 (below: Thoroughbred to the left, Palomino to the right).

The main difference of Thoroughbred except new technical process is, perhaps, crystal dimensions: the square is 80 sq. mm, more than by third less than 129 sq. mm of present Athlon XP, and also 116 sq. mm of Intel's Northwood also expected after New year.

There were dramatic changes of AMD's roadmap'2002: now, for example, the development of future 32-bit Duron line (Appaloosa core) is marked "according to market demand" from the second half-year 2002. The same with Athlon XP line (Barton), but from the beginning of 2003. However, roadmap of 64-bit Hammer processors are still firm, giving hope that closer to summer 2002 we shall see first prototypes of these models. AMD's plans to release mobile versions of Hammer were announced for the first time in November, however, they are still foggy without any clear terms.

Quite real mobile processors were announced as well: 1.2 GHz Mobile Athlon 4 and 950 MHz Mobile Duron. Both support AMD PowerNow! energy-saving technology and 3DNow! Professional, both are 0.18 micron. Systems with them are expected in December.

Without fanfares and sensations VIA gets closer to 1 GHz and higher chips: C3 866A (866 MHz, FSB 133 MHz, 1.35 V), C3 750A (750 MHz, FSB 100 MHz) and C3 700A (700 MHz, FSB 100 MHz) appeared for sale in Japan this month, and sales of C3 933 MHz (we've mentioned it in our review) are expected in the near future.

Indoubtedly VIA will become a competitor to present leaders, Intel and AMD, only on releasing the next generation of processors on C5N, C5X (C5XL and C5YL), C5W cores, with "normal-sized" cache, optimized pipelines and clock frequencies of 1 GHz and higher. Nevertheless, the significant progress of this company past year does not allow to speak of VIA as of insignificant market player as before. Having earned good sums on sales of economic models with a low cost price and, accordingly, price primarily in the markets of less developed countries, VIA will be able to afford considerable R&D expenses that should have an effect on productivity of future generations of processors.

There are just rumors about new Apple's processors. They say the company has already presented partners new Power Mac G5 systems with PowerPC 8500 processors. Maybe with 1.2, 1.4, and 1.6 GHz clock frequencies. New personal computers are informed to have FSB 400 MHz. However, I'll repeat: officially development of Apple's personal computer processors line will be more likely announced at MacWorld Tokyo in the beginning of the next year.

Transmeta... Continuous disappointments. Crusoe TM5800 notebooks are postponed again, now to the end of December. A many-month delay with release TM5800, that should have appeared in June, already worth Transmeta's CEO Mark Allen his post a month ago. I do not know about patience of notebook manufacturers, but more conservative companies releasing server systems, have started to turn away from Transmeta last month and, probably, will eventually completely turn to Intel's chips: Crusoe's prospects in economic servers are very good, but even remarkable "paper" CPUs won't work in such systems. Quite a real opinion of analysts: in one/two months Transmeta's authority might be fully undermined.

A pleasant announcement from Sun Microsystems: in the beginning of the next year the line of UltraSparc III processors will be expanded with a new 1.05 GHz model. Having the same architecture a new processor will have some differences from its predecessors, in particular, due to copper connections and low-k dielectrics of the new generation 1.05 GHz version will be manufactured on the same 0.15 micron technology, as the most productive at present 900 MHz UltraSparc III.

Some news about the specialized processors. In November Fujitsu licensed 32-bit RISC core from ARM. It's intended for system-on-a-chip (SoC) systems for future PDAs, phones, digital cameras, etc. Fujitsu licensed synthesized ARM926EJ-S core, known to be compatible with many operating systems and environments, and even more advanced ARM946E-S core. Fujitsu products are expected in the second/third quarter of 2002.

Toshiba has announced development of 64-bit TX4925 built-in microprocessor on the basis of MIPS architecture. The clock rate of the new processor is 200 MHz, core supply voltage - 1.5 V, I/O circuits voltage 3.3 V. Also: 32-bit data bus, 16 KBytes data cache; flash memory (such as NAND and Micron's SynchFlash technology) support, and 32-bit built-in PCI controller.

At last, Motorola specified its roadmap for the new generation of DragonBall processors. Now MPEG 4 video playback will be standard for Palm-like PDAs on the basis of DragonBall MX1. Motorola licensed MPEG 4 codec from Finnish Hantro, and now it will be added to ARM-like MX1. Having licensed ARM technology, Motorola is, certainly, interested in fast transition from the present 680x0 DragonBall manufacture to a new platform, even just to be equal to StrongARM processors and X-Scale from Intel expected in the near future.

STMicroelectronics has also announced a new chip. This time 166 MHz ST40RA166 processor on 32-bit SuperH SH-4 RISC core intended for systems with Windows CE 3.0 OS. Perfomance and functionality of this processor command respect: 300 MIPS, 1.2 GFLOPS, PCI, 64-bit DDR SDRAM interfaces supports, 32-bit peripheral bus. But, most likely, ST40RA166 will be used in set top boxes instead of PDAs.


Comdex Fall 2001 in Las Vegas became the main November event to bring a load of details about new motherboards and chipsets. News&details aplenty, so to say. The amount of chipsets Impresses and pleases, though it will, most likely, dazzle beginners wanting to choose a suitable motherboard. On the other hand, competition, increased in this sector, will favorably affect prices and will force chipset manufacturers to work harder on each new version.

November brought a wave of Intel i845D announcements, still informal (below is one of them: Gigabyte's GA-8IRX).

As it's known, official release of i845D is expected only on December, 14. i845D (or i845 B-step, as it's called now) will be delivered for the present i845 price of $40-42, and will likely drop the rating of SDRAM version showing insufficient productivity for considerable money. In addition, latest rumors are there's a variant to ship i845 B-step DDR chipset to motherboard manufacturers for a favourable price packaged with PC2100 DDR SDRAM memory modules (presumably from Nanya or Samsung). This again confirms Intel's firm intension to restore its status quo in Pentium 4 chipset market: SiS and ALi with licenses for releasing such production and VIA, still struggling for a right to release it, have been very active lately.

Speaking about Intel's prospects of P4 chipsets development, I'd like to say the company has made another firm decision to use DDRII memory in the future. There also was an interesting message from Taiwan that integrated 845G chipset (Brookdale-G) will be available to manufacturers in February 2002 (mass release of motherboards is expected in April). The release of this chipset has been planned for the third quarter before. Probably, Intel decided to force events due to (or contrary to?) increased activity of competitors.

Taiwan SiS is a doubtless hero of November. Its SiS645 DDR333 chipset for Pentium 4 as well as the version with integrated SiS315 graphic core named SiS650 were gladly accepted by manufacturers due to a rather low cost and excellent characteristics.

Except effective P4 platform support, SiS confidently remains the leader among chipset manufacturers for AMD platform: own factories allow the company to offer popular SiS735 chipset for a wholesale price of $16, and the new one, SiS745 chipset with DDR333 memory support for nothing more than $17 per unit.

The next pleasant surprise for motherboard manufacturers is expected to be mass appearance of SiS740 chipset for $26-28 that will allow to greatly lower motherboard production costs due to ubiquitous SiS315 integrated graphic chip.

There was some fun as well: the whole month Internet was filled with rumors that SiS was going to release P4 chipset with RDRAM support. As a matter of fact, rumors have started after SiS and Rambus have officially announced extension of relations regarding RDRAM licensing. Originally, the agreement of 1994 defined only PC600 licensing, now SiS can work with PC800 as well. Still the question about releasing RDRAM chipset by SiS is open: it would be tempting enough for the company to release a chipset with 4-bank RDRAM support, as Intel has cancelled plans to release Tulloch. However, SiS's officials abstain from comments, project realization is a big question.

There were numerous announcements of dual-processor motherboards on new AMD760MPX chipset (AMD762+AMD768 bundle): EpOX's M762A, Tyan's Thunder K7X (S2468), ASUS's A7M266-D, ABIT's WA-2A, Gigabyte's GA-7PXDW, MSI's MS-6501. Motherboards still are not for sale, nevertheless, they are expected to appear by the year-end (below is ASUS's A7M266-D).

Comdex Fall 2001 brought a load of NVIDIA nForce motherboards as well: both on the "senior" nForce 420 chipset version (K7N420 Pro from MSI, A7N266 from ASUS, NV22 from Abit, GA-7NTX from Gigabyte), and on "lighter" nForce 220 with 64-bit IGP (below is A7N266 from ASUS).

Moreover, in the second half of the month they've even announced sales of complete nForce 420 systems with new AMD Athlon XP processors and DDR SDRAM memory. Pity, a high enough price and not so high performance of built-in nForce graphics (lower than GeForce MX400 cards), now makes nForce not so actual. Mass appearance of nForce motherboards is expected in December.

VIA's news was very interesting and versatile in November. Perhaps, the most interesting event was the announcement of the second motherboard brand - VIARAMA, and bringing order to VIA Platform Solutions Division (VPSD), to be engaged in P4X266 motherboards distribution.

Now all motherboards developed by VIA and produced at partners' factories, will be sold under VIA trademark. Motherboards developed and produced from VIA's permission will be marked by VIARAMA brand, companies will market and sell such motherboards themselves under the aegis of VPSD. The third group of P4X266 motherboards is already known: it is the result of independent production by companies not afraid of Intel's anger and making motherboards despite everything.

P4X266 motherboards with VIA brand have already begun to appear for sale. Perhaps, not in very large amounts, but one will be able to find them if needed. However, in November VIA officially announced the new version of this chipset - P4X266A, with improved memory and ATA/133 interface support (bundled with VIA VT8233A south bridge). We've expected something like that from VIA, but only with FSB 533 MHz support of course. As P4X266A version was announced only with FSB 400 MHz, there's an opinion that a version, such as P4X266B will be quite real to appear. Probably, there will be a certain price difference between motherboards on "old" P4X266 and new P4X266A, otherwise, P4X266 sales may essentially stop.

Another announcement of the month was the release of P4M266 integrated chipset (VIA ProSavageDDR P4M266) for Pentium 4 systems. P4M266 specifications are absolutely similar to P4X266, the former just has ProSavage8 integrated graphic core from S3 Graphics. Products on this chipset still are not for sale, but, some other companies have already decided to release P4M266 motherboards, for example, AOpen (MX47M), DFI (NX30 and NM30) and others.

There was a whole load of announcements concerning becoming rather popular KT266A chipset for Athlon processors. The majority of manufacturers released motherboards on this chipset in various modifications and form-factors. The chipset has quite good productivity, is compatible with all Socket A processors, including the latest Athlon XP+. Most likely, VIA may expect quite good dividends in this sector despite hard competition with SiS.

It is also interesting to watch progress of VIA's new motherboard form-factor called Mini-iTX (just 170X170 mm). The company officially announced the reference design of VIA VT6010 motherboard in Mini-iTX form-factor in November.

VT6010 based on VIA Apollo PLE133 chipset will be equipped with integrated VIA C3 line E processor with 800 MHz clock rates and higher; it will also feature integrated AGP graphic controller, audio with SPDIF 5.1 out, TV-out, LAN port, and will be intended for small-sized mass PCs, "infostation", "infoservers", and network products. At a certain support of Taiwan motherboard manufacturers, Mini-iTX may quite settle down: such a tiny motherboard with a C3 processor that doesn't actually require any big heat-sinks, offered for a floor price, may quite become a founder of a new mass small-sized PC generation.

Only November brought us mass deliveries of a new revision of ALi's MAGiK 1 step C DDR chipset for Athlon processors. With a lower price than that of VIA KT266A and comparable speed, this new chipset has already interested all motherboard manufacturers, that had had to work with the old chipset version before. ASUS, SOYO, Iwill, and Transcend are going to ship motherboards based on ALi MAGiK 1 step C with optimizations for new Athlon XP in the near future.

Deliveries of Aladdin-P4 DDR333 chipset for Pentium 4 motherboards were delayed as well, and there was information that the amount of chipsets produced for the month had been just about 100 thousands. A surprising indifference to a promising market of Pentium 4 systems. Was it due to impossibility to establish mass production of Aladdin-P4? Or due to any special future surprises? We can only say that first Aladdin-P4 motherboards are expected in December; and mass amounts of these motherboard will, most likely, hit counters only after New Year.

And, finally, some exotics. Comdex Fall 2001 brought a unique now dual-processor Gigabyte's GA-8IPXDR motherboard on the prototype of Plumas chipset that is known to appear in a fairly long time.

The motherboard supports two Socket 603 Intel Xeon processors, including 0.13 micron Prestonia expected in 2002. Intel Plumas MCH will support two-channel DDR memory interface (the presented version supports only DDR200).

Another exotic exhibit was AT31 Socket A "concept" motherboard on ATi A3 chipset, presented by FIC.

A3 chipset is equipped with RADEON integrated graphic core and supports DDR SDRAM memory (PC2100/PC1600). According to some information, A3 is expected already in the beginning of 2002. The information that ALi and VIA chips will be used as A3's south bridge has also proved to be true.


The two main memory market themes of November were unconditional sharp price increase for separate memory products in the middle of the month, and continued fab stops along with tendencies of merging of this industry companies.

Increased prices for SDRAM / DDR SDRAM products in the second week of the month became a bit unexpected. Some manufacturers had even temporarily cancelled placing of orders until the wave of first days confusion ceased.

According to some sources, Micron and Samsung, failed to take out from Hynix with dumping pressure, have cancelled such policy and started negotiations at last that has somewhat allowed to control ship volumes. However, there was no panic on the market, and after a short week increase prices started to "dip" again.

It's, certainly, early to speak about final stability in this field. We've traced price changes in our daily news. As a whole, month results allow to say that a sharp price jump in the first half of the month was replaced by some reduction (though not to former positions), followed by a smooth growth. Decrease of chip production volumes on the one hand and market, rich enough with memory products, on the other hand have not found a golden middle yet.

All analysts speak very cautiously about further development of this situation. At least present DDR SDRAM prices increase caused by a wide spectrum of motherboards with DDR support, will be supported by i845D chipset release in the near future. Everything depends on whether manufacturers will have time to establish actual mass production of DDR SDRAM by this moment.

However, according to Elpida and Nanya representatives, ship volumes of DDR SDRAM will exceed those of usual SDRAM already in the second half of the next year. DDR motherboards, according to DRAM Exchange analysts, will take up to 60% of market in the near future already. Many manufacturers prepare for a wave of demand for this kind of memory hoping to replenish money at least this way. Samsung, Nanya, Hynix, and Micron have already announced that by the year-end they are going to increase DDR chip manufacture volume up to 15, 16, 7, and 2 million units per month, accordingly.

Preparations for DDR chips and modules of new generation - DDR333 (PC2700) became a thing of special interest, as support of this kind of memory is a constant attribute of latest chipset announcements. In November Hynix reported that its 128 MBytes and 256 MBytes PC2700 Unbuffered modules on DDR333 chips have successfully passed tests at Taiwan SiS, and Samsung has informed about getting a similar certification from Intel. Nanya's DDR333 business wasn't bad as well: during the first half of November the company shipped more than 120 thousands of chips that had successfully passed tests at leading Taiwan motherboard manufacturers - ASUS, Gigabyte, MSI, Elitegroup.

Not all memory manufacturers are so optimistic about prospects of their business. Here are some facts, reflecting current trends of DRAM market:

  • Past month NEC informed about reducing capacities of its factory in Roseville, USA manufacturing 200 mm semiconductor plates by two third. Now only 14 thousands plates per month remain from manufacture volumes of one of the largest USA DRAM factories (there have been up to 39 thousand plates before). Full termination of USA DRAM production and transition to LSI chips manufacture is at hand. It is interesting that once this factory have been thought of as NEC's first 300 mm manufacture.
  • Toshiba tries to solve the situation by a temporal (for some days in November and December) DRAM manufacture stop. Expected large losses of semiconductor division push the company to continue negotiations with Infineon about merging DRAM business.
  • Winbond has already reduced DRAM share in total production volume down to 30-40%. As it has become known, the company is going to become a new player in the multimedia chip market with its own developments.
  • Fujitsu is also going to change priorities and to turn to developing software, providing services, just keeping to flash-memory. Right at the end of the month it has announced termination of memory manufacture fab in Grasham, Oregon, USA; factory will be closed, staff laid off, and assets liquidated by the end of January, 2002. Now the company consolidates manufacture at Fujitsu-AMD Semiconductor Limited (FASL) factory in Japan.
  • Powerchip, having quite good modern factories, also leaves DRAM production, primarily turning to contract manufacture like TSMC and UMC. Nevertheless, right at the end of the month the company has announced additional increase of DDR chip production.
  • Elpida Memory plans to increase manufacture of specific memory kinds for workstations, servers, low-voltage memory for mobile phones and such.
  • Mosel Vitelic searches for partners to merge manufactures. If it succeeds, there's a high probability Infineon will be this partner. Companies already have Promos joint venture, however, Mosel Vitelic now insists on complete merge.
  • Hitachi again had to half manufacture volumes of 64 MBit and 256 MBit DRAM chips at its Singapore Hitachi Nippon Steel Semiconductor (HNS) factory to 10000 plates per month. Now manufacture load is only 30%. The further factory prospect is manufacture re-equipment for flash memory for various Hitachi devices along with SDRAM.

Maybe it's not a very joy image of November. However, even in the summer analysts warned that hard competition in the memory market would reduce the amount of players in this industry sector and consolidate remaining industrial forces. So, we watch this right now.

Comdex Fall 2001 brought us prototypes of SanDisk's 1 GBytes Compact Flash cards of Type I form-factor along with 1 Gbit NAND flash memory chips a joint development of Toshiba and SanDisk for these huge cards.

Chips will be produced on the same 0.16 micron technical process as present 512 Mbit ones. Their basis is the patented multi-level cell (MLC) technology developed by SanDisk, allowing to record two data bits to one memory cell, thus doubling its capacity. Two such modules can be placed in one TSOP package that enables to make 2 Gbit or 256 MBytes chips. Both companies will independently release CompactFlash, MultiMediaCard, and Secure Digital cards on the basis of these chips and will start selling them already in the first half of the next year.

Samsung announced new efficient 256 MBits NAND flash memory chips in November.

They feature 16-bit I/O bus, 1.8 V power supply, and will be produced according to 0.15 micron technical process.

Sharp and Winbond have united efforts to create 128 MBit and 256 MBit flash memory chips of the new generation based on technology Sharp ACT1 (Advanced Contactless Technology). Both companies will engage in development of 0.18 micron and 0.13-micron technologies for ACT1 flash memory production, Winbond will also manufacture and sell complete devices. ACT1 technology, developed by Sharp six years ago, allows to omit contacts in cells of NOR flash memory, enabling to reduce cell size almost twice. It is expected that inexpensive and capacious enough memory will be produced as a result.

Another curious November event: the list of companies, licensed long storage Ferroelectric Random Access Memory (FRAM) from Ramtron - Infineon, Fujitsu, Hitachi, Toshiba, Rohm, Samsung, Asahi Chemical, and recent Texas Instruments, was updated with NEC past month. FRAM popularity can be explained very simple: unlike other variants this memory nicely builts into integrated chips raising their cost price unsignificantly.

Write a comment below. No registration needed!

Article navigation:

blog comments powered by Disqus

  Most Popular Reviews More    RSS  

AMD Phenom II X4 955, Phenom II X4 960T, Phenom II X6 1075T, and Intel Pentium G2120, Core i3-3220, Core i5-3330 Processors

Comparing old, cheap solutions from AMD with new, budget offerings from Intel.
February 1, 2013 · Processor Roundups

Inno3D GeForce GTX 670 iChill, Inno3D GeForce GTX 660 Ti Graphics Cards

A couple of mid-range adapters with original cooling systems.
January 30, 2013 · Video cards: NVIDIA GPUs

Creative Sound Blaster X-Fi Surround 5.1

An external X-Fi solution in tests.
September 9, 2008 · Sound Cards

AMD FX-8350 Processor

The first worthwhile Piledriver CPU.
September 11, 2012 · Processors: AMD

Consumed Power, Energy Consumption: Ivy Bridge vs. Sandy Bridge

Trying out the new method.
September 18, 2012 · Processors: Intel
  Latest Reviews More    RSS  

i3DSpeed, September 2013

Retested all graphics cards with the new drivers.
Oct 18, 2013 · 3Digests

i3DSpeed, August 2013

Added new benchmarks: BioShock Infinite and Metro: Last Light.
Sep 06, 2013 · 3Digests

i3DSpeed, July 2013

Added the test results of NVIDIA GeForce GTX 760 and AMD Radeon HD 7730.
Aug 05, 2013 · 3Digests

Gainward GeForce GTX 650 Ti BOOST 2GB Golden Sample Graphics Card

An excellent hybrid of GeForce GTX 650 Ti and GeForce GTX 660.
Jun 24, 2013 · Video cards: NVIDIA GPUs

i3DSpeed, May 2013

Added the test results of NVIDIA GeForce GTX 770/780.
Jun 03, 2013 · 3Digests
  Latest News More    RSS  

Platform  ·  Video  ·  Multimedia  ·  Mobile  ·  Other  ||  About us & Privacy policy  ·  Twitter  ·  Facebook

Copyright © Byrds Research & Publishing, Ltd., 1997–2011. All rights reserved.