Apple Corp. recent purchased an old 200mm-diameter silicon wafer fab in San Jose capable of creating as small as 90nm device features. Formerly owned and operated by Maxim, the US$18.2M purchase reportedly includes nearly 200 working fab tools. Some people outside the industry have speculated that Apple might use this fab to do R&D on the A10 or other advanced logic chips, but this old tool-set is completely incapable of working on <45nm device features so it’s useless for logic R&D.
As reported at EETimes, this old fab could be used for the R&D of “mixed-signal devices, MEMS and image sensors and for work on packaging.” Those who know do not speak, while those who speak do not know…I do not know so I’m free to join the public speculation. Mixed-signal and MEMS processing would require major re-tooling of the line, but this 15-20 year-old tool-set is nearly turn-key for wafer-level packaging (WLP). With minimal re-tooling, this line could produce through-silicon vias (TSV) or through-mold vias (TMV) as part of Fan-Out WLP (FO-WLP).
Our friends at ChipWorks have published a detailed tear-down analysis of the System-in-Package (SiP) used in the first generation Apple Watch; it contains 30 ICs and many discretes connected by a 4-layer printed circuit board (PCB). Significant power and performance improvements in mobile devices derive from stacking chips in such dense packages, and even greater improvements can found in replacing the PCB with a silicon interposer. With Apple pushing the limits on integrating new functionalities into all manner of mobile devices, it would be strategic to invest in WLP R&D in support of application-specific SiP design.
A secretive investment holding company out of Hong Kong named GAE Ltd has acquired 98% of the shares in Silex Microsystems AB (Jarfalla, Sweden). The transaction took place on July 13th of this year when the former major shareholders agreed to sell all of their respective holdings, while Silex founder and CEO Edvard Kalvesten retains 2% of the shares in the company and continues his role as CEO and board member of Silex. No changes are made to the organizational structure or business operations of Silex, while the new owners plan to build a new high-volume manufacturing line near Beijing that clones the equipment and processes in Sweden with first wafers out by mid-2017 (as reported at EETimes).
Silex claims to be the “world’s number one Pure Play MEMS Foundry”, has worked with AMFitzgerald&Assoc. on RocketMEMS shuttle wafers to reduce MEMS development time by 6-12 months, and has developed multiple Through-Silicon Via (TSV) technologies to allow for efficient 3D integration of MEMS and CMOS.
Almost lost as a footnote in the news is that Silex holds IP on lead-zirconium-titanate (PZT) thin-film technology that allows for efficient piezo-electric energy-harvesting chips. MicroGen Systems is currently in the market with aluminum-nitride (AlN) piezo-cantilever micro-power generator system to power IoT nodes by scavenging either single-frequency or multi-frequency vibrations, working with X-Fab in Germany as foundry partner. If PZT-based piezo-cantilever energy harvesters can compete with AlN-based devices then the former could constitute much of the product volume in the new Silex Beijing fab. In 2014, Yole Developpement forecast “the integration of IoT-dedicated electronic components to result in a market volume of 2B units for these devices by 2021;” if 30% will use energy harvesting then this represents 600M units globally.
As reported in more detail at Solid State Technology, during the IEEE IITC now happening in Grenoble, imec and Lam showed a new Electroless Deposition (ELD) cobalt (Co) process that is claimed to provide void-free bottoms-up pre-filling of vias and contacts. The unit-process is intended to be integrated into flows to produce scaled interconnects for logic and DRAM ICs at the 7nm node and below. Co-incidentally at IITC this year, imec and Lam also presented on a new ELD copper (Cu) process for micron-plus-scale through-silicon vias (TSV).
The bulk resistivities of metals commonly used in IC fabrication are as follows (E-8 Ω⋅m):
Cu – 1.70,
Al – 2.74,
W – 5.3, and
Co – 5.8.
Of course, the above values for bulk materials assume minimal influence of grain sizes and boundary layers. However, in scaled on-chip interconnect structures using in today’s advanced ICs, the resistivity is dominated by grain-boundaries and interfacial materials. Consequently, the resistivity of vias in 7nm node and beyond interconnects may be similar for Cu and Co depending upon the grain-sizes and barrier layers.
The melting temperatures of these metals are as follows (°C):
Al – 660,
Cu – 1084,
Co – 1495, and
W – 3400.
With higher melting temperature compared to Cu, Co contacts/plugs would provide some of the thermal stability of W to allow for easier integration of transistors and interconnects. Seemingly, the main reason to use Co instead of W is that the latter requires CVD processing that intrinsically does not allow for bottom-up deposition.
Increasing fab costs coming for inspection and metrology At SEMICON West this year in Thursday morning’s Yield Breakfast sponsored by Entegris, top executives from Qualcomm, GlobalFoundries, and Applied Materials discussed the challenges to achieving profitable fab yield for atomic-scale devices (Figure source is the ITRS 2013 Yield Chapter). Due to the sensitive nature of the topic, recording was not allowed and copies of the presentations could not be shared. Qualcomm – Geoffrey Yu
Double-patterning will be needed for metal and via layers as we go before 90nm pitch for the next generations of ICs. Qualcomm is committed to designing IC with smaller features, but not all companies may need to do so. Fab costs keep going up for atomic-scale devices…and there are tough trade-offs that must be made, including possibly relaxing reliability requirements. “Depending on the region. If you’re in an emerging region maybe the reliability requirements won’t be as high,” said Yu. Through-Silicon Vias (TSV) will eventually be used to stack IC layers, but they add cost and will only be used when performance cannot be met with cheaper solutions. “An early idea was to use TSV for logic:memory,” reminded Yu, “but then there was innovation to LPDDR4 allowing it deliver the same bandwidth with one-half the power of LPDDR3, which delayed TSV.” GlobalFoundries – Harry Levenson
“A more expensive part could provide a better value proposition for a customer,” reminded Levenson as he discussed the challenges of inspecting next-generation commercial ICs in high-volume manufacturing (HVM). “We still have clear demand for products to run in HVM at the leading edge, but we are now in the world of double-patterning and this applies to optical inspection as well as imaging.” Requirements for inspection and imaging are different, but he same physics applies. In imaging Depth of Focus (DoF) of ~140nm is generally preferred, while the same used for inspection of a <140nm thin film would to induce noise from lower-levels. We can’t do e-beam inspections due to too much energy concentration needed to get acceptable throughput (and the challenge gets worse as the pixel area is reduced, inherently slowing down throughput). However, e-beams are helpful because they can detect open contracts/vias in metal levels due to the conductivity of electrons providing additional contrast compared to any possible optical inspection. Applied Materials – Sanjiv Mittal
Mittal discussed how the CMOS transistor gate formation process has increased in complexity over the last few device generations: 8x more unit-process steps, 3x higher fab cost, 50x lower defects needed for yield. “The challenges are immense,” admitted Mittal. “What happens when you try to work on yield improvement when you’re ramping volume? At the same time you’re trying to improve yield by making changes, you’re trying to increase the volume by not making changes.” Entegris – Jim O’Neill
O’Neill is CTO of the combined Entegris post-merger with ATMI, and was recently director of advanced process R&D for IBM. Since Entegris provides materials and sub-systems, in the simplest cases the company works to improve IC fab yield by minimizing defects. “However, the role of the materials-supplier should change,” averred O’Neill. “The industry needs bottle-to-nozzle wet chemistry solutions, and applications-based clean gas delivery.” In an exclusive interview with SST/SemiMD, O’Neill provided as example of a ‘wetted process solution’ a post-CMP-clean optimized through tuning of the brush polymer composition with the cleaning chemistry. ITRS Difficult Challenges for Yield 2013-2020
Existing techniques trade-off throughput for sensitivity, but at expected defect levels, both throughput and sensitivity are necessary for statistical validity.
Reduction of inspection costs and increase of throughput is crucial in view of CoO.
Detection of line roughness due to process variation.
Electrical and physical failure analysis for killer defects at high capture rate, high throughput and high precision.
Reduction of background noise from detection units and samples to improve the sensitivity of systems.
Improvement of signal to noise ratio to delineate defect from process variation.
Where does process variation stop and defect start?