TDEB Top Bar

Trending
Semiconductor Industry News — The Digital Electronics Blog
Follow Us

TDEB Masthead

The Semiconductor Hub

Semiconductors.  Chips and Systems.  Innovation.

Indexed by leading Search Engines, AI and referenced by Engineers worldwide 500+ Articles 10M+ Pageviews 30+ Reports 50+ Citations

Is Intel Preparing for 450mm Wafers? (2010)

Murugavel Ganesan
by
0
Share Facebook X LinkedIn WhatsApp Email

Is Intel preparing to push the issue of moving to 450mm wafers? One analyst thinks so. While most equipment is nowhere near ready (development has been limited), C.J. Muse, an analyst with Barclays Capital, thinks the No. 1 chipmaker could specify that its newly announced D1X development fab in Oregon will be 450mm capable, meaning it can accommodate 450mm processing when, or if, the equipment becomes available.

Considering that Intel, Samsung and TSMC have said in the past that they want to see 450mm development fabs by 2012 (fat chance), Intel almost has no choice but to build D1X with enough space to accommodate 450mm processing. While the chances of the tools being ready for 450mm by that date are slim, Intel must at least plan for the contingency if the company wants to send the message to its equipment suppliers that it is serious about this. While the 450mm ball is rolling to some extent (450mm wafer specs are ready; handlers and interfaces may be also), the question is when the big gear like lithography, CVD, RTP, etch systems and others will be ready. Don't be surprised to see Intel make a public statement about this fab being 450mm capable as a way to turn up the heat and spur tool development by equipment vendors who have mostly been dragging their feet. [via: EETimes]

Why wafer size matters

Chips are made many at a time on round slices of silicon called wafers. In 2010 the most advanced fabs used 300mm wafers. A 450mm wafer has 2.25 times the area of a 300mm wafer, so each pass through a process tool produces more than twice as many chips. If a tool for 450mm wafers costs much less than 2.25 times as much as its 300mm version, the cost per chip falls. That is the whole argument for moving to a bigger wafer.

Why it was hard

Every tool in the fab, from lithography scanners to etch, deposition, cleaning, metrology and the robots that carry wafers, has to be redesigned for the bigger wafer. Equipment makers carry most of that development cost, and only a few chipmakers can afford to build the fabs that would buy the tools. That is the stand-off the post describes: chipmakers wanted the tools, and tool makers wanted proof that chipmakers would buy them.

What "450mm ready" meant

In the EE Times report, Muse drew a careful line: his checks suggested D1X would be "450-mm ready" but not "450-mm capable". In practice that meant a building prepared for bigger tools, with taller ceilings, more clean-room air handling and floor pedestals able to carry heavier tools and contain their vibration, without any 450mm tools inside yet. He described it as Intel keeping its options open.

Deep dive: the numbers and details

This part holds the geometry, the readiness checklist from 2010 and what happened after this post. Skip it if the overview is all you need.

Area and dies per wafer

Wafer area grows with the square of the diameter: (450 / 300) squared = 2.25. The number of whole dies grows slightly faster, because the partial dies lost at the round edge are a smaller share of a bigger wafer. A common approximation for gross dies per wafer is:

DPW = pi x (d/2)^2 / S - pi x d / sqrt(2 x S)

where d is the wafer diameter and S is the die area. For a 100 mm² die:

WaferArea termEdge loss termGross dies
300mm70767About 640
450mm1,590100About 1,490

That is about 2.33 times as many dies for 2.25 times the area. Yield, scribe lines and edge exclusion change the exact numbers but not the trend.

The readiness list in 2010

The EE Times report listed what Muse and Sematech counted as done or not done for 450mm at the time:

ReadyNot yet done
Requirement guidelines; early design; early prototypes; interoperability test bed; mechanical wafer bank; technology intercept node defined; single crystal wafer bank; equipment performance metricsMetrology and process equipment development; equipment prototypes; equipment demos; actual equipment readiness

The report also noted that equipment vendors were reluctant to build 450mm tools because of the cost, that only a handful of chipmakers would build 450mm fabs, and that Intel, TSMC and Samsung were the ones pushing.

What happened next

  • 2008 background. As The Register recounted in 2022, Intel, Samsung and TSMC announced in May 2008 that they would cooperate on a move to 450mm wafers, and the Global 450mm Consortium (G450C) was later formed with Intel, TSMC, Samsung, IBM and GlobalFoundries to develop the equipment.
  • 2013. The same article reports a former TSMC executive's account that TSMC turned away from 450mm in 2013, and that the industry focus moved to advanced process technology instead.
  • 2014. Semiconductor Engineering reported in July 2014 that 450mm was on hold, with major toolmakers and foundries stepping back to reassess, and that even Intel agreed it was on hold until the end of the decade.
  • 2016 and 2017. EE Times Asia reported in October 2017 that G450C had quietly finished the first phase of its work at the end of 2016 with no date set for a next phase, leaving 300mm fabs to carry the load.

So the analyst was right that Intel would prepare for 450mm, but the wider bet did not pay off: the industry stayed on 300mm wafers.

Why the economics did not close

The geometry gives at most about 2.3 times the dies per wafer pass. For 450mm to pay, tool costs had to rise much less than that, and the industry also had to fund years of tool development for a handful of buyers. The reports above show that the chipmakers' priorities moved to new process nodes instead, and without broad buyer commitment the tool makers had no reason to finish the work.

Murugavel Ganesan
About the Author
Murugavel Ganesan
Product Director | SerDes · IVN · Edge AI · Compute | Ex Intel · Infineon

A semiconductor product leader working across product architecture, definition, strategy, and silicon and software execution, from concept through volume production. He has worked with teams across Asia, Europe and the USA, including at Intel, Infineon and ArrayComm.

His work covers ASICs, SoCs, FPGAs and chiplets for wireless, wireline, automotive, IoT and AI, including PCIe, CXL and DDR interfaces alongside RF and mixed-signal design.

Share Facebook X LinkedIn WhatsApp Email

Post a Comment

0 Comments

Your comments will be moderated before it appears here.

Post a Comment (0)

#buttons=(Ok, Go it!) #days=(20)

Our website uses cookies to enhance your experience. Learn more
Ok, Go it!