Pete's Posts Blog

Should lifetime of EUV optics be a concern?

It’s well known that EUV adoption is running later than hoped, mostly due to inadequate source power (although ASML and Cymer say they are on track to provide workable solutions and imec says it’s on track for the 10nm node). After that, the main challenge could be those associated with EUV mask blanks, which are essentially sophisticated mirrors. The dual challenge there is that they are not only difficult to produce without defects, but they are difficult to inspect. Presently, the only way to really test them is to fabricate them and see what kind of pattern results after they’re used.

But another challenge recently came to my attention: the optics in the EUV system, which are also sophisticated mirrors made of multi-layer structure, get contaminated during operation. This degrades their quality over time, and eventually the system must be disassembled and the optics recoated or replaced.

EUV optics F1

I recently talked to Dr. Harro Hagedorn, head of R&D at Leybold Optics in Alzenau, Germany. Located about 20 minutes outside of Frankfurt, the company supplies evaporation and magnetron sputtering systems used to fabricate the multi-layer coatings used for EUV collector optics and or many other applications such as synchrotron labs and X-ray devices. And they work with ASML and Zeiss, and the Fraunhofer Institute.

Speaking on EUV, Dr. Hagedron, said the output of the light source is still not high enough, and also the lifetime of the optics was a concern. “This light source is normally an awful thing for the optics because you have a metal droplet that is heated up by a laser and then it creates a plasma. These metal droplets are also contaminating the optics,” he said. To correct, this “they have to disassemble the system and recoat these optics. They are very expensive. Also, the life throughput that comes from this light source and through the optics goes down. They have to also manage this,” he said.

Part of the complexity and expense of the optics is that they rely on interference coatings that require stacks of layers. “The challenge is that these layer stacks are incredibly thin, 3-4nm, with coating uniformities in the range of 0.1%,” Hagedron said. “It’s not any more than a diameter of an atom.” The goal for these optics is a reflectivity of nearly 70%.

Investigating a bit further, I found that there has been a significant amount of research into the lifetime of EUV optics. In fact, earlier this year in April, a session at SPIE was dedicated to damage to VUV, EUV, and X-ray Optics. One of the papers by Laser-Lab in Germany, KLA-Tencor and Fraunhofer, described work that characterized EUV damage thresholds and imaging performance of Mo/Si multilayer mirrors. Here’s a summary:

Currently, more and more powerful EUV sources for next generation semiconductor microlithography are being developed, for which novel optical elements like multilayer or grazing-incidence mirrors are required. Consisting of very thin alternating layers, especially molybdenum and silicon for the wavelength of 13.5 nm, multilayer mirrors are typically employed for near-normal reflection angles. These mirrors are presently being optimized with respect to thermal resistivity and reflectivity. However, only very few ablation and damage threshold studies at a wavelength of 13.5 nm are available up to now for these optical elements.

We studied 1-on-1 and 10-on-1 damage thresholds of Mo/Si multilayers with EUV radiation of 13.5 nm wavelength, using a table-top laser produced plasma source based on solid gold as target material. The experiments were performed on different types of Mo/Si mirror, showing no significant difference in single pulse damage thresholds. However, the damage threshold for ten pulses is ≈60 % lower than the single pulse threshold, implying a defect dominated damage process.

Using Nomarski (DIC) and atomic force microscopy (AFM) we analysed the damage morphologies, indicating a primarily thermally induced damage mechanism for higher fluences. Additionally, we characterised transmission and reflection properties of novel Mo/Si multilayer beam splitters performing wavefront measurements with a Hartmann sensor at 13.5 nm wavelength. Such wavefront measurements allow also actinic investigations of thermal lens effects on EUV optics.

My main takeaway from all of this is that even if the technical challenges of EUV make it ready for production for the 10nm device generation, there still must be lots of questions about the ultimate cost of ownership and how that will compare to double and triple patterning approaches with 193nm immersion.

What’s down the road for bulk FinFETs

For the 10nm node and beyond, transistor research efforts are focused on high mobility designs with Ge and III-V channel, reducing VDD supply voltage as well as the subthreshold slope in transistors and optimizing multi-Vt  designs. Eventually, lateral finFETs built from silicon nanowires may be required. As previously reported in the post “Status update on logic and memory roadmaps,” the 14nm node (which imec calls the “N” node”) is in development today, heading toward early production in 2013/2014. That will be followed by the N10 node in production at the end of 2015 and beginning of 2016. Then N7 and N5 will follow in 2017 and 2019.

A detailed look at the likely roadmap for logic devices built on bulk silicon wafers using finFET technology was provided at the recent International Technology Forum for the press at imec in Leuven, Belgium in October. An Steegen, senior vice president process technology provided the overview, highlighting research underway for the 10nm, 7nm and 5nm nodes.

Steegen F1

Steegen said power was a concern in both high performance logic devices, which are thermally limited, and in mobile devices, which are battery limited. “What we’ve been trying to do at all our technology nodes is to try to step down that power curve, mainly be trying to lower the Vdd,” she said. The trick, she said, was to lower the power, but still retain performance, and the best way to do that is to make the subthreshold slope of the device steeper. She said the target was 16mv/decade, which is the limit of conventional transistors.

Steegen F2

There’s a tradeoff, however, in that reduced Vdd often means increased variability, depending on the threshold voltage of the devices. “On the High Vt device, when you go low Vdd, you’re so close to the threshold voltage of your device that the spread becomes enormous. What you could do is switch to a low Vt device. That’s better for variability but your leakage is going to increase,” she said.

Steegen F3

At imec, their research is focused on bulk silicon finFETs (others are exploring fully depleted SOI) with a replacement gate and high k. Work is still underway on ways to best integrate a replacement metal gate, and on multi-Vt devices. “That’s still work we are executing,” she said. She noted that imec has worked on high-k metal (HKMG) gates for more than 15 years, and is now looking at how to implement a replacement metal gate on a finFET device and enginner the Vts. “Uou also need to make sure your reliability comes together,” she said.

One way in which they have enabled multi-VT tuning (range up to 600mV) is with controlled Al diffusion in metal gate stack. “You want to make sure you can offer the designers a low Vt device, a standard Vt device and a high Vt device, which means that you need to be able to tweak the materials in your gate to cover that entire range of Vt scaling,” Steegen said.

Steegen F4

She added that, for the 10nm node, they are engineering an entire silicon finFET platform. “That means we work on every single module going in, from scaling to the N10 dimensions, control of the fin height, making sure you get conformal doping in the fin, and source/drain engineering because you still want to get some form of stress from your source/drain.”

Steegen F5

For the 7nm node, Steegen believes the channel materials will need to be replaced with higher mobility materials, germanium for PFETs and InGaAs for NFETS. To integrate these materials, a technique known as Aspect Ratio Trapping (ART) is used. This technique uses high aspect ratio sub-micron trenches to trap threading dislocations, greatly reducing the dislocation density of lattice mismatched materials grown on silicon. “You have engineer your dislocations and defects,” Steegen said.

ART is shown to be very effective for a wide variety of materials including Ge, GaAs and InP. It has been combined with epitaxial lateral overgrowth to create long, 18 micron wide strips of low dislocation density material. ART has been used to integrate many types of Ge and III-V devices on silicon including GaAs MOSFETs, GaAs lasers, GaAs tunnel diodes and a silicon infrared imager chip with monolithically integrated Ge photodiodes.

For PFETs, the technique involved, “recessing the silicon, growing silicon germanium buffer back and then strained germanium on the top. The STI is then going to be recessed and you have a strained germanium fin on the top,” Steegen explained. The same integration scheme is used for NFETs, but “it’s a little more complex to try to get to a strained InGaAs NFET channel because the lattice mismatch with silicon is larger. You have to use more buffers here and go a little bit deeper to grow all these buffers through the trench,” she said. “Aspect ratio trapping makes sure all the defects — and you’re going to have them in that strained/relaxed buffer — are trapped at the sidewalls of the STI so that they don’t reach the top silicon.”

Steegen F6Steegen F7

This added complexity appears to be worth the effort based on modeling, which shows a net gain of 25% more performance at constant power, compared to a bulk silicon finFETs. “There is still a lot of benefit you’re going to get in one node by replacing these channel materials,” Steegen noted.

Steegen F8

Beyond 7nm, imec is looking at higher mobility materials such as graphene, and also looking at new device architectures such as tunnel FETS. “At this point, we are looking at germanium source tunnel FET to overcome the tunneling barrier with a lower bandgap material at the source,” Steegen said. “We truly want to try to break that 60mv/decade subthreshold slope.” She said lots of progress has been made but there was more work to do to understand band-to-band tunneling mechanisms. The team is also looking at “2D” materials such molybdenum disulfide and tungsten diselenide.

Steegen F9Steegen F10Steegen F11

EUV is late but on the way for 10nm; DSA is promising

EUV lithography is late, but it is on the way and will be ready for insertion into the 10nm node, which is slated to go into production in late 2015/early 2016. Meanwhile, results from early work into directed self-assembly (DSA) is quite promising. DSA could be used in conjunction with EUV for the 7nm node, scheduled to go into production in the 2017/2018 timeframe. These were some of the conclusions from the imec International Technology Forum for the press earlier this month, where the latest results from EUV and DSA work were presented. Imec and ASML also announced an advanced patterning center that will be based at imec’s Leuven campus focused on EUV.

Luc Van den hove, president and CEO of imec, described EUV as a cost-effective lithography approach that is “absolutely needed.” He said: “We realize that EUV is late. There are challenges here. But I have to say over the last couple of months, significant and steady progress has been realized.” In terms of imaging performance, imec has been characterizing some of the latest hardware together with ASML and have showed very good resolution performance of 13nm half pitch and 22nm contact holes. “With double patterning, we have even demonstrated 9nm half pitch,” Van den hove said. “Who would have thought a couple of years ago that this would be realizable with lithography?”

EUV F1.1

Adequate source power, which directly determines throughput, is one of the challenges. “We are seeing a steady increase quarter by quarter,” Van den hove said. “We see that we get the improvements and I’m very convinced that very soon EUV will be ready to enter manufacturing.”

An Steegen, senior vice president of process technology at imec, provided some more details: “The standalone Cymer source has been demonstrated at 55Watt. The next checkpoint is the 80W by the end of 2013. That is on the assembled system, source and scanner, and [throughput] should be approximately 58 wafers/hour. The goal is by the end of 2014, about 126 wafer/hour,” she said.


Steegen said ASML’s 3300 system has already been verified. “We definitely and clearly can see the resolution benefits as well as overlay capability,” she said (the demonstrations were done in at ASML’s facilities in Veldhoven; imec has a 3000 system installed, but won’t get a 3300 until February 2014).

The ideal entry point for EUV is the 10nm node (or N10 using imec’s terminology). “If you look at the cost calculation, the best entry point for EUV is actually at N10 because you can replace triple patterning layers in immersion with a single patterning layer in EUV,” Steegen said. Since that will come relatively soon with early production occurring toward the end of 2015 and in early 2016, that means that likely the whole development phase will have already been built on immersion and multi-patterning. “Likely you will see on the most difficult levels, a swap, an introduction of EUV at the most critical levels later on in manufacturing for N10,” Steegen said. “That is still what the forecast is today. You would also see a benefit later on inserting EUV in N10.”

When you move to N7 and do multi-patterning, it’s getting even more complicated, since almost every level is going to be triple patterning. “If you replace that with EUV, will still are going to try to have single patterning on most of the levels, but there’s some complexity coming in even with EUV, that you would have to go to multi-patterning for N7 to get to those dimensions,” Steegen added.

However, she acknowledged that more work needed to be done.

She noted that defectivity in mask and the reticles was still a challenge, particularly those defects embedded in the multi-layered films. “Between those mutli-layers in the reticle, you might have embedded defects that you can’t see today with any inspection tool. You actually need an actinic inspection tool that also runs at the wavelength of the EUV tool to really see those embedded defects on the reticle. Today, the only way we find those is by printing the wafer and see the printable defects in your patterns,” she said.  

Van den hove added: “Now that the tool is really progressing tremendously and becoming ready for manufacturing, you also have to make sure that the rest of the infrastructure is ready. In our program, we are now focusing very much also on how to handle defects – defectivity on the mask is one of the big challenges.”

In an earlier interview, Veeco’s Tim Pratt, Senior Director, Marketing, said that indeed the next major roadblock to progress in the ongoing push to develop EUV lithography for volume production is the availability of defect-free mask blanks. He said that the tools in place today are not capable of producing mask blanks with the kind of yield necessary to support a ramp in EUV.

Imec is also evaluating the use of a pellicle on EUV reticles. Pellicles are used to keep particles from falling on the reticle during exposure and transport, but since EUV masks are reflective instead of transparent, it’s no clear how a pellicle would work. “For years, we have been thinking that pellicles would be impossible in EUV, and whether we can use a pellicle or not,” Van den hove said. But he said there were some options that are being evaluated at this moment. Steegen said: “We’re looking into would it make sense to avoid added defects during scanning to introduce a pellicle on these EUV reticles.”

Steegen also said they have also seen good progress with the resists needed for EUV for the 22/20nm contact hole and line space range, including improved line edge roughness (LER) and local critical dimension uniformity (LCDU). “We see quite some improvement in LER post-etch and also in LCDU post-etch where we combined the exposures on the 3300 with etch and basically try to improve our line edge roughness and local CDU,” she said. Imec has integrated litho and etch together in such a way that the resulting dimensions are improved after etch. “What you can see is that you cannot only shrink your holes after etch, but you also improve your line edge roughness and local CDU after etch,” Steegen said. “That’s pretty significant because basically you can shrink the nominal dimension of the hole by about 14% and the local variation of that CD, we can improve about 30% post etch.”

Here’s how imec summed up EUV’s readiness:  

• NXE-3300 resolution benefit and overlay capability demonstrated.

• CH and LS resist materials for NXE-3300 selected, based on CDU, LER/LCDU and defectivity; good progress in resist process down-selection towards 16nm LS (@ 0.33NA)

• 4.2nm of CH CD reduction for 26 nm HP post etch (14%); 1.4nm LCDU reduction for 26nm HP post etch (30%)

• Source power and mask defectivity remain key challenges:

• ASML/Cymer demonstrated 55W power on standalone Cymer source, outlook 80W (58wph) by YE13,

250W (126wph) by YE14.

• Introduction of actinic inspection tools by 2015 for mask blank embedded defect detection and of pellicle to reduce mask defect adders.

• Expectation that EUV will be introduced at a few critical levels in N10 nanufacturing (replace LE3) with cost reduction benefit. Potential EUV area benefit from tip-to-tip and tip-to-line and pitch scaling, with redesign.

DSA is very promising

Van den hove described direct self-assembly (DSA) as “very promising” and Steegen said work there has largely focused on reducing defectivity. In DSA, resists that contain block copolymers are deposited on top of guiding structures. The self-directed nature of the process results in very regular patterns with very high resolution.


The trick with DSA is that it requires a double exposure to take away the random patterns at the edge of the device, and the resolution needed for this “cut mask” is also very high. “We’re convinced that it’s not a replacement for EUV or any high resolution lithography technique. We are very convinced it will be used in conjunction with EUV,” Van den hove said. “It certainly keeps the pressure on EUV very high.”

Steegen described DSA as a complimentary litho technique that is having quite some momentum. The process starts with a “relaxed” guiding pattern on your wafer.  Then, depending on the polymer length in the block copolymer, the space in between the guiding structure is replicated into multiple lines and spaces. “The defectivity of these materials are going to be key to bring the defects down. Our year end target is 60 defects/cm2 and this needs to go down even further next year,” she said.


Work at imec has shown that the polymers, with a hard mask on top, are robust enough to enable the etching of the patterns into silicon. “That’s fairly new data and very promising,” Steegen said. Imec is already looking at where DSA levels could be inserted into the logic N7 flow, with fins and spacers being primary targets. Steegen said the Metal1 level would be a challenge due to its irregular pattern. “That makes it not easy to be replaced with DSA, but we’re looking into techniques to do that,” she said.

Here’s how imec summed up DSA readiness:

• Good progress in material selection and integration flow optimization for line-multiplication down to 14nm, pattern transfer into bulk Si demonstrated.

• First templated DSA process available using SOG/SOC hard mask stack.

• Focus on defectivity reduction & understanding, currently at 350 defects/cm2, YE13 target 60 def/cm2

• Alignment and overlay strategy needs to be worked out

• First N7 implementation levels identified: Finfet (replace SADP EUV or SAQP 193i) and Via (replace EUV SP/DP or 193i LE3).

Intel stands firm on 450mm; challenged by defects at 14nm

Brian Krzanich, chief executive officer of Intel, said Intel is standing firm on 450mm development (despite rumors to the contrary), during a quarterly conference call with financial analysts. He also blamed defects on a slight push-out of next-generation 14nm technology.

When asked about 450mm plans, Krzanich said: “We have not changed our timing. We are still targeting the second, latter half of this decade. We continue to see great value in 450. It brings tremendous economic value to everybody who participates in it. We continue to work with our partners. We are here part of the joint development program in New York, continuing to work on 450. We continue to work with our partners, especially TSMC and Samsung and we are still targeting the back half of this decade. This is a long 10-year program when you really take a look at it. So I think you will get mixed signals throughout those 10 years,” he said.

As noted in my last post, progress is on track at the G450C consortia — an initiative by five big chip makers, Intel, TSMC, GLOBALFOUNDRIES, IBM and Samsung, partnered with New York state and CNSE — to develop 10nm capability on 450mm wafers in 2015 or 2016.

Krzanich also commented on the status of the firm’s 14nm roll out. “We continue to make progress with the industry’s first 14nm manufacturing process and our second generation 3D transistors. Broadwell, the first product on 14nm, is up and running as we demonstrated at Intel Developer Forum, last month. While we are comfortable with where we are at with yields, from a timing standpoint, we are about a quarter behind our projections. As a result, we are now planning to begin production in the first quarter of next year,” he said.  

When asked about why the delay, Krzanich said it was “simply a defect density issue,” and said it was just part of the development process. “As we develop these technologies, what you are doing? You are continually improving the defect densities,” he said. “As you insert a set of fixes in groups, you will put four or five, maybe sometimes six or seven fixes into a process and group it together, run it through and you will expect an improvement rate occasionally as you go through that,” he said. He said the fixes don’t deliver all of the improvements. “We had one of those,” he said. “Why do I have confidence? Because, we have got back now and added additional fixes, gotten back onto that curve, so we have confidence that the problem is fixed, because we have actually data and defects and so that gives us the confidence that we are to keep moving forward now.”

Intel has already started construction on a 450mm pilot line at its Ronler Acres location on Northwest Highway in Hillsboro. The D1X module 2 is about the same size (1.1 million square feet) as the original fab D1X and is built specifically for 450mm wafers. When the second module is complete, it will start up on 450mm wafers once it is equipped with appropriate manufacturing tools and gear.

Intel is currently equipping its D1X development fab to process 300mm wafers using 14nm manufacturing and expects to initiate production this year. While the D1X module 1 facility is 450mm-capable, it will come online as a 300mm fab.

Image by Portland Business Journal.

Image by Portland Business Journal.

Intel DX1

TSMC is the only other company with an effort underway to develop a 450mm production facility. In June 2012 it was reported that the Taiwanese government had approved a proposal to build a 450mm wafer fab in central Taiwan early in 2014. TSMC said back in 2011 that it planned to install 450mm pilot lines within a couple of established wafer fabs — Fab 12 in Hsinchu and Fab 15 in Taichung, Taiwan.


Progress on 450mm at G450C

At Semicon Europa last week, Paul Farrar, general manager of G450C, provided an update on the consortium’s progress in demonstrating 450mm process capability. He said 25 tools will be installed in the Albany cleanroom by the end of 2013, progress has been made on notchless wafers with a 1.5mm edge exclusion zone, they have seen significant progress in wafer quality, and automation and wafer carriers are working.

G450C is an initiative by five big chip makers — Intel, TSMC, GLOBALFOUNDRIES, IBM and Samsung – partnered with New York state and CNSE. The main goal is to develop 10nm capability on 450mm wafers in 2015 or 2016. “What we have to demonstrate is that a film on 300mm, when we scale it up to 450mm, we can do it with the same capability and, more importantly, at a very significantly reduced cost per process area. In other words $/cm2 need to go down significantly. That’s how you hit the scaling that we’ve typically seen in a wafer transition which is in the 30% range,” Farrar said.  

G450C aims to develop 10nm capability on 450mm wafers in 2015/2016.

G450C aims to develop 10nm capability on 450mm wafers in 2015/2016.

Farrar said the facility looks quite different now than it did in March, when it was fairly empty. 18 tools have been installed so far, with a total of 25 tools delivered into the Albany complex by the end of 2013. “2013 is the year that I call install and debug,” Farrar said. “We’ll have approximately 50% of the toolset in the facility by the year end. It doesn’t mean that they’ll all be up and running but they will be placed in Albany or virtually at the suppliers, with about 35% of the toolset coming in 2014 and the last little bit that will be delivered will be the lithography tool in early 2015.” The program is organized around unit processes, including: film deposition and growth, wafer clean and strip, CMP and other processes, inspection and metrology, etch and plasma strip, and lithography.

In call cases, G450C will have at least one process that will be required for the 14nm flow. In most cases (about 70%) they will have multiple suppliers, at least two and sometimes three. “At the end, we’ll have both unit process and what I would call modules – 2 or three step processes – demonstrated. And then our member companies will take those building blocks and they will put their devices and their IP and then go build out factories,” Farrar said.

Farrar showed data demonstrating significant progress in wafer quality. He noted that they now have one wafer supplier and a second one coming on line. He also said automation and carriers were working well. “I don’t think they’ll be showstoppers. There are always things you can learn but those are working reasonably well,” he said.

G450C is also trying to take advantage of having a clean slate to make a switch from notched wafers – which provide a useful indicator regarding the crystal orientation of the silicon – to notchless wafers, which are perfect circles. “If you think about the physics around a notch, it really makes it difficult to get uniform films,” Farrar said. “A circle is a lower stress form. We get 1-1.5% better in getting closer to the edge. Using chips around the notch and perhaps getting to 1.5mm edge exclusion. We won’t get there if we don’t have notchless wafers. Our goal is to collaborate with our IC makers, our tool suppliers and materials suppliers, along with our facilities group.”

Probably the most critical part of the 450mm puzzle is lithography. Farrar said the consortia has been working with Nikon. “We were able to work with Nikon so that we now have immersion capability, in Japan, starting in June of 2014 and we’ll then have that tool installed in Albany at the end of the first quarter of 2015. We will have a true lithography capability which will enable us to get the efficient and actual process recipes that the deposition supplier will need to see so that they can demonstrate the capabilities at the 450 wafer form factor,” he said. “In the interim, we’re working on DSA (directed self assembly). We’re starting to see some pretty good results. I don’t think this will be a high volume technique but it’s a way that we can get something that works started in the early process modules in 2013 and early 2014.”

Wafer quality has improved, and wafer reclaim efforts are underway. “When we started this program, we had a handful of wafers. That was in the 2012 timeframe. We started to get reasonable test and monitor wafers in late 2012, and if you look at where we are today, in the 2nd half of 2013, we have a quality spec where we’re hitting bout 98% of the area is in spec, and the particle level is effectively every wafer is meeting the specification. We still need a little work on wafer flatness,” Farrar said. The next step is what he called “prime” wafers, which they expect to have in the middle of 2014.  

Status update on logic and memory roadmaps

The way in which logic and memory technology is likely to evolve over the next six years was provided at imec’s recent International Technology Forum in Leuven, Belgium. An Steegan, senior vice president process technology at imec, said that FinFETs will likely become the logic technology of choice for the upcoming generations, with high mobility channels coming into play for the 7 and 5nm generation (2017 and 2019). In DRAM, the MIM capacitor will give way to the SST-MRAM. In NAND flash, 3D SONOS is expected to dominate for several generations; the outlook for RRAM remains cloudy.

As shown in the Table, the 14nm node (which imec calls the “N” node”) is in development today, heading toward early production in 2013/2014. That will be followed by the N10 node in production at the end of 2015 and beginning of 2016. Then N7 and N5 will follow in 2017 and 2019.

Imec's view on the future technology roadmap for logic and memory.

Imec’s view on the future technology roadmap for logic and memory.

The most notable evolution in the logic roadmap is that of device architecture, where planar devices are being replaced by fully depleted devices. There are two main flavors of fully depleted devices: fully depleted SOI (FDSOI) and finFETs.  Imec sees FDSOI as an option for 14nm, which is “actually a speed push option from 20nm,” Steegen said. “What’s happening is that in the 14-16 generation, speed push knobs are implemented on the technology roadmaps to get the extra performance boost for that node.” That’s partly driven by the readiness (or really unreadiness) of EUV.  “Scaling is not necessarily the .7X one dimensional scaling that you expect node to node,” Steegen said. That’s why, in the 16-14nm generation, planar devices are being replaced by a higher performing fully depleted device. “When you push this to 10 and 7nm that we believe that the finFETs are going to have a long lasting life,” she added, which means that we will see finFETs on the roadmap for at least three generations.

The two main advantage of fully depleted SOI versus planar: 1) area footprint. You always get more performance from a trigate device since you actually use that third dimension. 2) Power/performance benefits.  

Steegen said imec is now mainly focused on assessing processes for 7nm and trying to figure out when the ultimate finFET scaling limit will be hit. At that point, expect to see what imec calls “local SOI,” which is a slight undercut of the bulk silicon fin to provide better isolation in the well. A more extreme version gate all-around device, which could be based on silicon nanowires.

To boost performance in the past, external source/drain stressors were used to increase electron and hole mobility in the device. The problem moving forward, in the N10 and N7 generations, is that there’s no space to do that. Instead, expect to see replacement of the silicon channel with a high mobility material.  “When you look at what material that could be, germanium is a good candidate to push hole mobility, so the PFET. And III-V, InGaAs, is a good material for NFET devices to push the electron hole mobility,” Steegen explained.

As far as standalone memory (vs embedded memory) goes, STTRAM is now being pushed forward to basically replace the MIMCAP on the DRAM roadmap. That’s because it’s very challenging to get an EOT of 0.3 (see table) and maintain acceptable leakage of the MIM capacitor.

For NAND flash, Steegen said the two-dimensional hybrid floating gate integration flow is definitely being pushed to a 15 and 13nm half pitch. “Scaling is one challenge you’re going to encounter here. The other one is the charge you can trap on the floating gate itself. It becomes so discrete there’s hardly any charge left. The variability you’re going to have on the hybrid floating gate concept is likely getting too big. That is why 3D SONOS is definitely getting it’s way in on the NAND flash replacement roadmap and also we forecast that to have quite a long lifetime: two or three generations,” she said.  

Steegen said the outlook for RRAM was cloudy. It could be the eventual successor to 3D SONOS, but “if you want to replace 3D SONOS, you’re getting to need 3D RRAM because you’re going to use the same 3D configuration. Also, for NAND flash replacement, you always need that select element to make sure you only select that one cell you want to turn on. How to integrate the selector with the RRAM element in a 3D configuration is going to be the trick of how RRAM can enter this NAND flash roadmap at the end,” she said.

Semiconductors that detect cancer

At the recent imec International Technology Forum Press Gathering in Leuven, Belgium, imec CEO Luc Van den hove provided an update on blood cell sorting technology that combines semiconductor technology with microfluidics, imaging and high speed data processing to detect tumorous cancer cells.

As I reported previously, the challenge is huge: one has to have the ability to detect one bad tumor cell in 5 billion blood cells. This equate to a requirement to detect 20 million cells per second.

“We all know that cancer is one of the most severe illnesses, creating a tremendous burden on the patients, on families and on society,” Van den hove said. Cancer spreads throughout our body through circulating tumor cells that originate from the primary tumors, and create secondary tumors. Usually those are the most fatal ones. “If we can develop a system that can detect those circulating tumor cells in a very early phase, we develop an early warning system for cancer,” he added.

Today, single cell analysis requires a lot of manual operations, sophisticated (and expensive) tools that require a lot of time to generate the results. Very often those results are not accurate enough or not sensitive enough. “What we actually need is a much more engineering type approach where we start from the clinical samples directly, and one that is much more automated. At imec, we are developing a high content, high throughput cell sorter, which is much more compact than any cell sorter ever made. We have demonstrated the proof of principle of operation of such a cell sorter,” Van den hove said.

Here’s how it works: When a blood sample is loaded into the system, the cells flow in via the main microfluidic channel. The sample contains a whole range of cells. Several types of white blood cells, red blood cells, platelets, and even maybe some rare blood cells, such as circulating tumor cells. The microfluidic channel is about 30 microns wide, about the size of a single cell. Conditioned light illuminates the cells and results in fringe patterns that are recorded by the CMOS image sensor. These fringe patterns contain information on the 3D structure of the cell. The recorded holograms are then reconstructed into 2D images that are used to distinguish between the different types of cells based on cell size and nuclear morphology. These calculations, managed by an ASIC, need to be performed very quickly before the cells arrive at the microheaters further down the channel. The microheaters located at the crossings of outlets, generate small and short-lived steam bubbles that gently but quickly push single cells to a particular outlet. While the red bloods cells just go straight on, the while blood cells are brought to the lower outlet and the tumor cells are pushed in the upper outlet. The sorted cells can then be further analyzed downstream. For instance, by extracting specific molecular information. After sorting, high resolution 2D or even 3D imaging of each single cell can be consulted for further visual analysis.

Cell sorter F2

Imec has so far demonstrated all the building blocks of this technology and demonstrated proof of principle with a single channel. The next step is to build thousands of channels on a single chip. “With silicon technology, we can very easily integrate thousands of those channels on one chip and in this way, realize the enormous sensitivity that is needed in order to detect these bad tumor cells in a billion cells. The system we’re developing will be able to process more than 20 million images a second. This kind of tool will bring one to bring this kind of analysis from a very sophisticated lab to the side of our bed and provide much better accuracy,” Van den hove said.

Cell sorter F1

Europe’s 10/100/20 program

Despite some economic woes in recent years, Europe remains dedicated to building a strong electronics industry. This was brought home to me recently when, in advance of Semicon Europa (October 7-10 in Dresden), I had a chance to talk with Heinz Kundert, president of SEMI Europe. “There are several initiatives like the KET (key enabling technology) initiative that are working on the same goals to increase the competitiveness but also to get more manufacturing back to Europe,” he said. One of these is the Horizon 2020 effort, the EU’s new program for research and innovation is part of the drive to create new growth and jobs in Europe, which will run from 2014 to 2020 with a budget of just over €70 billion (some announcement during Europa is planned). France also announced Nano 2017 and plans to invest approximately €3.5 billion in France up to 2017 in nanoelectronics.

Another interesting project, specifically aimed at boosting semiconductor manufacturing in Europe is the 10/100/20 program, which has a goal of generating €10 Billion in public/private funding for R&D, €100 billion euros investment for manufacturing, and 20% share of global chip production market by 2020. Neelie Kroes, European Commission Vice-President, commented in May of this year: “I want to double our chip production to around 20% of global production. It’s a realistic goal if we channel our investments properly. A rapid and strong coordination of public investment at EU, Member State and regional level is needed to ensure that transformation.”

Already, some significant progress has been made. Five new pilot lines were launched in May 2013 under the ENIAC Joint Undertaking (EU public-private funding program), worth over €700 million and bringing together over 120 partners. These pilot lines allow research centers and companies to cooperate across borders to test and perfect new technologies and tools, such as: technologies and equipment for GaN-based substrates; 450mm equipment and materials; 300 mm power semiconductors; new MEMS materials and packaging; and 28/20nm FD SOI.

In the past, Europe was home to around 17% of global semiconductor manufacturing, but that has declined to around 6-7% today. Turning that around to reach 20% in the next seven years will be a challenge, but where there’s a will there’s a way. “Although there are a lot of issues, and a lot of details that need to be clarified – many people are questioning whether it’s possible or not possible — but what I see is positive thinking and this is most important,” Kundert said.

If you plan to visit Semicon Europa and want to learn more, you’re in luck. “We will address all these issues on public funding, public/private partnership vision at Semicon,” Kundert said. There will be an executive summit this is addressing that issue, as well as sessions on funding to explain how SMEs (equipment suppliers) can participate. “Projects are going to be presented so people understand what’s behind all these initiatives and visions and big numbers,” he added.

Join us for a free webcast on Advanced Packaging, Sept. 30th

Please join us for a free webcast on Advanced Packaging, to be held on September 30th, at 11:00am Eastern, 10:00am Central and 8:00am Pacific. You can register in advance at this link:

The webcast, which will feature presenters from Texas Instruments and Micron, will address how today’s packaging technology is driven by a combination of cost, performance, form factor and reliability. The presenters will examine new advances in conventional back-end packaging, including wafer bumping and copper wire bonding, as well as the role of new 2.5D and 3D integration. They will also focus on issues related to cost, performance (speed, power and noise immunity), form factor (thickness, weight, PCB area consumption), and testability, as well as the tradeoff of technical maturity versus risk in high-volume manufacturing. The webcast will also include new information on the Hybrid Memory Cube, a three-dimensional structure with a logic device at its base and a plurality of DRAMs vertically stacked above it using through-silicon via (TSV) connections. This is a revolution in 3D integration packaging technology, as covered in our September issue cover story.


Our first speaker, will be Dr. Mahadevan “Devan” Iyer. As Director of TI’s Worldwide Semiconductor Packaging operations, Dr. Iyer oversees a global team that drives a process to determine the packaging design and technologies that best meet the requirements of our customers in measures of miniaturization, performance cycle time, and cost. Dr. Iyer joined TI in 2008 to lead the global SC Packaging team. He has more than 25 years of experience in the microelectronics industry. Dr. Iyer is a recognized authority in semiconductor packaging technologies.  He has more than 150 technical publications and 28 patents to his credit.

Iyer_Devan_11-20-12 (199x300)

Our second speaker will be Aron Lunde, a Product Program Manager for Micron’s Hybrid Memory Cube.  His responsibilities include coordinating Micron’s internal departments to develop, construct, and deliver the HMC to meet the demands of multiple business engagements. Mr. Lunde joined Micron in 1994 and worked as a Test Engineer for Micron’s Boise, Lehi and Singapore facilities.  He later worked as a Mobile DRAM Designer, implementing repair and test modes on original Micron designs, integrating repair schemes into DRAM and mobile device architectures, and developing tools to evaluate the efficiency of repair schemes. Mr. Lunde has a BS in Electrical Engineering from the University of Idaho and has issued 15 patents.


Thanks to our webcast provider, TalkPoint, you’ll be able to tune in using any mobile device including iPads, tablets and phones!

The webcast will be archived after the event and can be accessed for 12 months. Register now at

Defect-free mask blanks next EUV challenge

The next major roadblock to progress in the ongoing push to develop EUV lithography for volume production is the availability of defect-free mask blanks. According to Veeco’s Tim Pratt, Senior Director, Marketing, the tools in place today are not capable of producing mask blanks with the kind of yield necessary to support a ramp in EUV. “Based on the yield today, the mask blank manufacturing capacity can’t produce enough mask blanks to support the ASML scanners that they’re planning to ship,” Pratt said. “ASML is going to be delivering some light source upgrades in the field and when those start happening, the effective total wafer throughput of EUV scanners in the field is going to multiply and there’s just not the supply of usable mask blanks to be able to support those.”

The requirement for 2015 is to have zero blank defects larger than 62nm. SEMATECH in 2012 reported work showing eight defects larger than 50nm. “A lot of progress being made but the elusive zero defects has not yet been hit,” Pratt said. Veeco, which is the sole supplier of EUV multilayer deposition tools, has plans to upgrade the existing Odyssey tool and launch a new platform in the 2017/2018 timeframe.

Figure 1 shows an EUV mask, which is considerably more complicated than conventional photomasks. The EUV mask begins with a substrate. On the back of the substrate you have some material that’s used for chucking (an electrostatic chuck is used to hold the mask to a stage in the ASML tool and in the Veeco ion beam deposition tool). On top of the substrate is a multilayer sandwich made up of 40-50 moly silicon pairs that creates a mirror. A ruthenium capping layer helps protect the mask. The top layer is an absorber, and that’s what gets patterned.

Click to view full screen.

Click to view full screen.

The photo at the bottom right of Fig. 1 shows a small pit on the substrate. “As the multilayer gets deposited on top of it, you take what in the beginning might have been a small pit and at the top it becomes 1.5X or so larger,” Pratt said. “This a common problem with EUV. These very small pits and bumps on the substrate, because of the deposition angles, the way that the multilayer is put down, as small non-killer defect on the substrate suddenly becomes a killer after deposition.”

The left photo is a larger particle that fell on the blank during deposition. Pratt said that even if you could mill that down and make it level, you would just never get the reflectivity out of the section that you need.

Where is EUV today? Billions have already been invested to build the EUV infrastructure with particular emphasis on the light source. Chipmakers have invested in ASML, and ASML acquired light-source provider Cymer. There has also been a very large Industry investment in Zeiss to build the AIMS tool, which is a defect detection and repair system at EUV wavelengths.

In July, ASML said NXE:3300 scanner imaging and overlay performance reached levels where they are engaging with customers on a strategy for the 10nm logic node insertion (23nm half pitch). Good imaging performance was shown down to 13nm half pitch, and overlay between the NXE:3300 and NXT systems, had been demonstrated at less than 3.5nm. Good performance, stability and reliability of the pre-pulse source concept was demonstrated with a rate of around 40 wafers per hour, and ASML expressed confidence in reaching the goal of 70 wafers per hour productivity in 2014.

What could derail the EUV ramp, according to Pratt, is a supply of defect-free mask blanks. “EUV is, despite many years and many dollars of investment, not yet in production. The two main gaps are the EUV light sources and the defects on the mask. As they start to make progress, people start to look more seriously at the list of things to worry about for EUV going to production. Number one on that list is the mask defects. Mask defects can come from all different sources during the entire process, from the substrate all the way through to usage in the fab,” Pratt said. “The most dangerous (un-repairable) defects come from the ML (multilayer) coating process during mask blank manufacturing. You can’t clean them and you can’t repair them and if you have more than some very small amount, there’s really nothing you can do about it. You just have to throw that mask blank away and try again, which creates a very large selling price for the mask blanks. Not just because they are difficult to make but you’re throwing away a substantial amount of what you’re trying to sell,” Pratt said.

Figure 2 illustrates the process flow for a EUV mask blank. After a substrate polishing process, the substrates goes from the substrate supplier to the mask blank supplier. At the mask blank supplier, they will deposit the multilayer, the fiducial mark and the deposition. The blank gets sold to the mask shop, which could be either captive or a merchant. That blank, which is basically a mirror at the point, gets patterned and inspected and sent off to the fab. Pratt said that once the mask hits the mask shop, there is a little more leeway in terms of the defects because the defects that occur in the mask show are usually on top. “It’s usually absorber type defects or patterning type defects and those are a lot more easily repairable,” he said.

Click to view full screen.

Click to view full screen.

Figure 3 shows the timeline of Veeco’s system developed, starting with a research system developed in 1996 that went to Lawrence Berkeley. The was optimized in the 2003-2010 timeframe, which included work with SEMATECH in a joint development program. That basically turned it into what Pratt describes as an R&D system. “We have a system that is being used for all the mask blanks in the world. But those mask blanks are really R&D blanks that people use for print checks and reflectivity checking, but certainly nothing they would use in a fab yet, or expect to get yield off of,” he said. “A lot of the time, you don’t know if it’s yielding or not until the very end of the process.”

Click to view full screen.

Click to view full screen.

Pratt said they have seen some improvements when it comes to defects. “We’re not yet where we need to be for logic high volume manufacturing, but we’re getting close to where we need to be for memory.” The real issue is the low yield. “At the current yields, that mask blank makers would need to spend a whole lot of money, probably on the order of $3 billion or so, on capital to meet what the mask blank demand would be over the next five years. That’s just not feasible. EUV clearly can’t ramp in that scenario,” Pratt said.

Veeco is addressing the defect challenge in two ways. The short-term solution is an Odyssey upgrade. The longterm solution is a new platform. “The Odyssey upgrade improves the yield of the tool. But then longer term we think the next gen is needed, especially as you get out to years 4 and 5 where high volume manufacturing starts to occur,” Pratt said.

Veeco F4 new

The ion beam deposition system in shown in Figure 4. The target assembly rotates, so the process might start off with silicon, the assembly is then rotated to deposit molybdenum and rotated again to deposit ruthenium. The problem is that the ion beam doesn’t always direct hit the target. “You might have some of these high energy ion missing the target and hitting the chamber. The chamber has shields on it, but that ion can bounce around and when it hits the shields, there’s a chance that it can knock off particles,” Pratt said.

The Odyssey upgrade will: reduce source to target ion overspray and reduce high energy reflected neutrals. New ion source optics are planned as well as a larger target size. Lower beam energy operation and lower pressure operation are also planned. Those should have two benefits.

Longer term, the next generation EUV ML system will focus on particles and CWL (center wavelength) process repeatability (CWL is a measure of how reflective the mask is). The new platform will minimize particle proximity, and accommodate new source technology. A larger chamber, out-of-plane deposition geometry, low-defect clamping and integrated endpoint control are also planned. Figure 5 shows progress in defect reduction from 2004 to 2012.

Click to view full screen.

Click to view full screen.

If you have some news to share, send me an e-mail at,