Profit Control
Wednesday 1st January 2003
ACCELERATED
SEMICONDUCTOR TECHNOLOGY ROADMAPS, FUELLED BY POWERFUL ECONOMIC
FORCES, HAVE CREATED THE PHENOMENON OF RAPIDLY SHRINKING PROCESS
WINDOWS. CONVERGENCE OF TODAYS ADVANCED PROCESS CONTROL,
ADVANCED EQUIPMENT CONTROL, FAULT DETECTION/CLASSIFICATION
AND PROCESS WINDOW MONITORING TECHNOLOGIES INTO A MORE GENERALISED
PROCESS WINDOW CONTROL (PWC) WILL ENABLE EVEN FASTER DESIGN-RULE
SHRINKS FOR BOTH LOGIC AND MEMORY MANUFACTURERS, WRITES KEVIN
M MONAHAN OF KLA-TENCOR.
Untitled Document
| | ACCELERATED SEMICONDUCTOR TECHNOLOGY ROADMAPS, FUELLED BY POWERFUL ECONOMIC FORCES, HAVE CREATED THE PHENOMENON OF RAPIDLY SHRINKING PROCESS WINDOWS. CONVERGENCE OF TODAYS ADVANCED PROCESS CONTROL, ADVANCED EQUIPMENT CONTROL, FAULT DETECTION/CLASSIFICATION AND PROCESS WINDOW MONITORING TECHNOLOGIES INTO A MORE GENERALISED PROCESS WINDOW CONTROL (PWC) WILL ENABLE EVEN FASTER DESIGN-RULE SHRINKS FOR BOTH LOGIC AND MEMORY MANUFACTURERS, WRITES KEVIN M MONAHAN OF KLA-TENCOR. |
| | At the 90 and 65nm technology nodes, most shrink and performance showstoppers will be caused by systematic parametric errors. Factory managers will be forced to make an economic trade-off between yield, density and performance in order to achieve maximum average selling prices (ASPs) at minimum die cost. Compounding the problem, widespread use of sub-wavelength low-k 1 lithography with resolution enhancement technologies is shrinking both the pattern registration and critical dimension process windows. In the case of lithography, systematic process errors appear in the form of measurable parameters, such as overlay (OL) or critical dimension (CD). Other parameters, including exposure variation and local defocus, can lead to systematic errors that are observed indirectly in the form of critical CD and OL excursions. | | Fig.1: APC requires a control model (e.g. litho controller) and a process model (e.g. CD versus dose). The control model can be more effective if CD measurement data is both accurate (e.g. error less than 1nm) and current (e.g. delivered in less than 1hr). The process model is more useful when the systematic model variation is large relative to the process noise contribution. Noise may derive from process window variation that is outside the context of the process model and therefore not correctable by APC | |
| The two generic means of controlling gate CD are advanced process control (APC) and process window management (PWM). APC seeks to adjust process parameters inside process windows. PWM seeks to align the process windows, primarily by enabling recurrent equipment matching. Both APC and PWM are evolutionary next-steps in the copy exact philosophy of semiconductor manufacturing. Advanced process control APC is a closed-loop control system in which process correctables are fed back to the preceding module or fed forward to the next to reduce parametric variation on product wafers. The success of APC is based on two assumptions: the existence of an effective control model and the existence of a valid process model (Figure 1). For an effective control model, timely measurement data must be supplied to allow high control gains to be used without danger of overcorrecting (Figure 2). For the process model to be valid, it must account for a useable fraction of the observed variation and leave a minimum of unexplained variation in the form of residuals. |
| Unexplained variation can have both systematic lack-of-fit and random repeatability components. Quantum effects notwithstanding, if we consider enough factors, nearly all the parametric variation in a semiconductor factory would be classified as systematic. The postulate of APC, however, is that control can be accomplished by adjustment of a few critical parameters inside otherwise stable, multivariate process windows.
The effectiveness of APC can be compromised by delays in either processing or measurement. Figure 2 shows simulated reduction of lot-level variation as a function of maximum allowable measurement delay. These control curves are similar regardless of the origin of the factory data used for simulation. In some cases, the effectiveness of CD APC can double by taking delays down to about 1 hour from typical values in the range of 5-10 hours. This behaviour is also observed for APC in overlay applications. For lot-level control, measurement delays can be reduced by increasing stand-alone metrology and compensation for incremental cost on every tool.
| Alternatively, in the case of stand-alone CD and overlay metrology, dramatic improvements in APC performance can be achieved through optimising capacity and sample plans. For example, increasing an overlay sample plan from four arbitrarily selected fields to ten statistically optimised fields may nearly double the potential effectiveness of OL APC. The requisite metrology capacity pays for itself by accelerating shrinks and enhancing profitability. | | Fig.2: Estimate of lot APC effectiveness versus time delay. About 90% of value is achieved for delays less than 1hr |
|
|
|
| Process window monitoring PWM is an open-loop control system in which multiple parameters are monitored for the purpose of detecting excursions and matching process equipment using monitor wafers. PWM is a powerful tool for dealing with unexplained intra-module or inter-module process variation that can reduce performance of APC, particularly, since performance gaps commonly arise from collapse and/or displacement of process windows. As an example, Figure 4 shows litho and etch CD windows as a function of litho defocus. The optimal litho and etch windows are displaced by 0.2µm, creating a 30% etch-CD risk and the potential for catastrophic yield loss. Most CD APC applications are blind to this class of inter-module variation. The root-cause of etch window displacement may not be the CDs themselves, but rather the resist sidewall angles as they change through focus and erode during etch. We expect that PWM will be most effective when embedded in smart metrology tools that provide large amounts of unique information. In addition to its role in supporting APC, PWM can be used by itself as a fully automated, web-based monitor for process tool stabilisation and tool matching across entire areas. In lithography, CD PWM is implemented using multiple focus-exposure matrices measured on a scanning electron microscope (SEM) or scatterometry (SCD) metrology tool (Figure 5). Automated analysis of these matrices can generate dose, focus and depth-of-focus metrics for each lithography cell and each critical layer in a factory. Figures 6, 7 and 8 provide examples of results from PWM for differing purposes. Figure 8 shows how PWM can exceed the capability of in-situ metrology. Combining flash memory and foundry data from two different factories, we observed a maximum focus tracking error of 30nm without calibration. Our development work indicates that, with calibration, the noise floor for CDSEM-PWM system is less than 12nm at 30. |
| Problem looms at 65nm Currently, CD shrink entitlement is being held back by poor correlation of traditional lot-based CD averages to device yield - and yield entitlement is falling. Shrinking design rules are narrowing the overall CD process window, creating a looming CD-related yield problem for the 65nm technology node. With the advent of 300mm wafers and 130nm design rules, yield-affecting CD control problems have shifted to the cross-wafer and cross-field levels. This problem has contributed to the delay of 130nm technology in the foundries. Going forward, the International Technology Roadmap for Semiconductors (ITRS 2001) has identified the 65nm technology generation as a showstopper for gate CD control, with no known solutions. Nevertheless, recent data has shown promising results for a scatterometry-based process window monitor used for cross-field CD control applications. Cross-field error is typically the largest systematic component of variation. Much of this is correctable with a large revenue impact. Calculations suggest that the weekly added value of correcting these errors on a microprocessor line with 1000 wafer starts per week could be as much as $6.0 million from shrinks, $5.8 million from speed enhancements and $2.9 million from extending the life of 200mm fabs. Many CD errors are caused by cross-field deviations in focus and dose (Figure 9). The long-term scanner focus variation, as measured by PWM, can have up to a 200nm range. By contrast, the long-term repeatability of the PWM focus measurement is less than 10nm, meeting the ITRS requirements for focus metrology at the 65nm node. Lost shrink and speed binning opportunities currently result in millions of dollars of lost revenue per year - and the situation will get worse at 65nm. These losses will be due primarily to shrinking process windows and persistent cross-field variation. Scanner metrology will no longer work at 65nm because focal tests have relatively poor measurement precision and fail to provide simultaneous dose and depth-of-focus information. |
| Fast scatterometry-based process window monitors with 10nm focal precision can provide a faster, cheaper, better solution. PWM can be moved to inexpensive metrology tools, liberating $20-30 million lithography cells for wafer production. | | Fig.3: Graphical representation of the 50nm lithography data. The largest single component of variation is associated with intra-field sites, suggesting that the order of economic value for APC is site, lot, field and then wafer. The composite estimates support the idea of intra-field dose control and raise some questions about the value of wafer-to-wafer adjustments. Note that much of the random variation is due to multiple systematic variations confounded by sparse sample plans | | | Fig.4: Litho-etch focus window interactions can
produce unexplained variations in CD APC |
|
| | Fig.5: SCD-based process window monitor with 10nm focal measurement precision suitable for the 65nm node. The system also works with eCD1 CDSEM technology |
|
|
Process control at 90nm and beyond For the 90nm technology node and beyond, we foresee a trend toward extended, generalised, factory-wide, model-based, APC. The traditional economic benefits of APC are many. These include reduced process variation, accelerated shrinks, elimination of send-ahead wafers, fewer monitor wafers, shorter response times, reduced scrap, better tool matching, improved overall equipment effectiveness, faster yield ramps, lower parametric yield loss, better device performance, easier process transfer from pilot line to factory and dramatically lower labour costs. These benefits, especially in the case of microprocessors, will continue to translate into higher gross margins for semiconductor products since they simultaneously lower manufacturing cost and increase average selling price. | |
| In addition, we expect traditional APC architectures to be augmented at the 90nm node to support the precursors of fully depleted silicon-on-insulator transistor technology such as Intels Terahertz project. At the 90nm node, physical gate lengths are expected to reach 50nm and decrease rapidly to 30, 20, and 15nm in subsequent generations. Some microprocessor manufacturers may fabricate 50nm gates at the 130nm node. Such design rules will generate more rigorous process control requirements for rapidly shrinking, overlapping process windows. Ultimately APC architectures may be called upon to provide comprehensive control of transistor formation. To address the new requirements, APC architectures must incorporate: - New
metrology tools, measurement targets and analysis software to ensure that accurate, yield-relevant corrections are fed to the APC system. Examples are CD scatterometry and next-generation overlay tools that use grating targets to reduce measurement uncertainty and improve device correlation.
- Infrastructure
extensions that enable seamless root-cause determination and correction when the APC history analysis generates an action request. Examples are model-based process window monitors (PWM), advanced equipment control (AEC), fault detection/classification (FDC), integrated diagnostic monitors, and internet-based support.
|
| - Factory-wide
open APC frameworks (such as KLA-Tencors Catalyst) that allow single-point manufacturing execution system (MES) integration and co-ordination of multiple APC applications running simultaneously in several process areas. Consider an L effective controller that feeds forward chemical mechanical planarisation (CMP) thickness data to correct CDs in litho, litho CD data to correct CDs in etch and etch CD data to correct L effective during spacer formation or ion implantation.
- Business
rule management that supports high-mix production and multiple APC strategies. The simple APC business rules developed initially in low-mix microprocessor and DRAM factories must be extended to address high-mix manufacturing contexts in foundries and to support multiple strategies that require different levels of metrology and process tool dedication for low and high ASP products.
- Multi-threaded
APC applications that respond to more than one input and generate more than one output. Consider a litho area that exhibits systematic cell-to-cell, lot-to-lot, wafer to-wafer, field-to-field and die-to-die CD variation that might be corrected by feeding back dose offsets for each cell, lot, wafer, field and die, respectively.
|
| - High-bandwidth
support for stand alone, clustered and integrated metrology tools. Stand-alone and 300mm cluster metrology tools generating data on a lot-to-lot basis will be joined by integrated metrology tools that generate data on a wafer-to wafer basis. In some cases, bandwidth requirements may force on-tool distillation of correctables required for lot-based recipe modification.
- Physical
model-based process control that can predict electrical characteristics. APC applications will evolve from simple offset corrections (CD control) and empirical response surfaces (overlay control) to physical models that predict electrical characteristics such as effective gate length (L-effective) and equivalent oxide thickness (EOT).
- Control
of transistor formation using models that predict device performance. Model-based process control will evolve from simple models to more sophisticated physical models using measurements from multiple tools to predict device performance characteristics such as voltage threshold (V-threshold), off current (I-off) and drive current (I-drive).
| | Fig.7: A large overlap window is critical for manufacturing. Here, PWM assists in determining the common lithographic focus-exposure window over five positions in the scan field. Overlap windows can be spatial, temporal, or tool-to-tool |
|
|
|
Ultimately, driven by compelling economic forces, todays APC, AEC, FDC and PWM will evolve into a new species of process control that we refer to as process window control. PWC opens a vast array of possibilities. Gate stack formation will take on new meaning as we force the overlap of multivariate process windows from deposition to alignment, exposure, etch, spacer formation and ion implantation. Intra-field proximity effects in lithography may be compensated for in the etch module by varying power, pressure, flow and chemistry using feedback from embedded measurement tools. Cross-wafer non uniformity induced in the etch module may be corrected in lithography using field to-field and intra-field dose control. Factors that were once ignored in APC, such as lithographic defocus or spacer sidewall angle, may become an integral part of process correction as scatterometry proliferates. APC, AEC, FDC and PWM will continue, but will function more robustly inside the context of process window control. | | Fig.8: Performance of the PWM system must exceed that of available in-situ metrology on process equipment. Flash and foundry data from two different companies indicate maximum error of about 30nm without calibration (~12nm with calibration) | Links to gate CD control align="justify"Gate CD control is an eminent example of how process window control can be used. Assuming constant yield, links can be established between gate CD control, die-density, average selling price and factory life extension. |
| Characterising the variability of gate CD can be used to separate out correctable systematic cell, lot, wafer, field, die and site variation from lot-to-lot temporal variation. Benefits accruing from improved CD control are contingent upon a number of assumptions, including a price performance premium, unsaturated market demand, scalability of interconnect speed, dimensional scalability of other critical layers and the relatively small incremental cost of PWM and APC. Simple calculations can establish an expected entitlement for maximum incremental return against which the actual return on investment (ROI) may be compared. The Pareto chart in Figure 10 shows such examples of incremental return due to microprocessor speed improvement, increase of die density, factory life extension and rework reduction assuming APC produces a 15% to 30% decrease in lot-to-lot CD variation. Since the model for total CD variation includes lot, wafer, field and site components, the actual shrink ratios are only about 0.95 and 0.90, respectively. Initial production is assumed to be 1000 wafers per week, with 150 dies per wafer at $100 per die. These entitlements assume that gate CD is the critical speed and shrink bottleneck. We are not modelling the interactive incremental return due to simultaneous increase of speed and density. Return on investment can be increased dramatically by controlling the other components of CD variation at the site, field and wafer levels (Figure 3). The extension of APC to more critical steps and layers could remove speed and shrink bottlenecks in other parts of the process. Finally, the effectiveness of the APC algorithm itself could be improved by making changes in the manufacturing context that minimise measurement delay and the contribution of unexplained process variation. |
| | Fig.9: Process windows define the ranges of dose and focus that create < 10% CD variation. Process windows will shrink at 65nm | | Fig.10: Pareto of predicted incremental revenue due to microprocessor speed improvement, increase of die density, factory life extension, and rework reduction | Some features of the future semiconductor production landscape emerge: - Economic
considerations will drive demand for more granularity in parametric control.
- Control
of module-to-module parametric window overlap will become a necessity as gate CD control merges with L-effective control.
- Far
more effective and complex APC applications will appear as measurement and process delay times are reduced.
- Further
reduction of unexplained variation will be accomplished using process window monitors, smart metrology tools and statistically optimised sample plans.
- Further
reduction of delays will be accomplished by increasing the level of metrology integration, clustering and dedication.
In difficult economic times, the metrology-driven shrink may be the most cost-effective alternative to massive investment in new process equipment or factories since it both extends the life of current assets and provides the foundation for higher returns during economic recovery. Advanced process window control is a key enabler for this strategy. |
|
|
|
AngelTech Live III: Join us on 12 April 2021!
AngelTech Live III will be broadcast on
12 April 2021, 10am BST, rebroadcast on 14 April (10am CTT) and 16 April (10am PST) and will feature online
versions of the market-leading physical events:
CS International and
PIC International PLUS a brand new
Silicon Semiconductor International Track!
Thanks to the great diversity of the semiconductor industry, we are always chasing new markets and developing a range of exciting technologies.
2021 is no different. Over the last few months interest in deep-UV LEDs has rocketed, due to its capability to disinfect and sanitise areas and combat Covid-19. We shall consider a roadmap for this device, along with technologies for boosting its output.
We shall also look at microLEDs, a display with many wonderful attributes, identifying processes for handling the mass transfer of tiny emitters that hold the key to commercialisation of this technology.
We shall also discuss electrification of transportation, underpinned by wide bandgap power electronics and supported by blue lasers that are ideal for processing copper.
Additional areas we will cover include the development of GaN ICs, to improve the reach of power electronics; the great strides that have been made with gallium oxide; and a look at new materials, such as cubic GaN and AlScN.
Having attracted 1500 delegates over the last 2 online summits, the 3rd event promises to be even bigger and better – with 3 interactive sessions over 1 day and will once again prove to be a key event across the semiconductor and photonic integrated circuits calendar.
So make sure you sign up today and discover the latest cutting edge developments across the compound semiconductor and integrated photonics value chain.
REGISTER FOR FREE
VIEW SESSIONS