Overlay process control up to and including the 45nm node has been implemented using a small number of large
measurement targets placed in the scribe lines surrounding each field. There is increasing concern that this scheme does
not provide sufficiently accurate information about the variation of overlay within the product area of the device.
These concerns have led to the development of new, smaller targets designed for inclusion within the device area of real
products [1,2]. The targets can be as small as 1-3μm on a side, which is small enough to permit their inclusion inside the
device pattern of many products. They are measured using a standard optical overlay tool, and then calibrated. However,
there is a tradeoff between total measurement uncertainty (TMU) and target size reduction . Also the calibration
scheme applied impacts TMU.
We report results from measurements of 3μm targets on 45nm production wafers at both develop and etch stages. An
advantage of these small targets is that at the etch stage they can readily be measured using a SEM, which provides a
method for verifying the accuracy of the measurements.
We show how the 3μm in-chip targets can be used to obtain detailed information for in-device overlay variability and to
maintain overlay control in successive process generations.
As overlay budgets continue to shrink, there is an increasing need to more fully characterize the tools used
to measure overlay. In a previous paper, it was shown how a single-layer Blossom overlay target could be
utilized to measure aberrations across the field of view of an overlay tool in an efficient and low-cost
manner. In this paper, we build upon this method, and discuss the results obtained, and experiences gained
in applying this method to a fleet of currently operational overlay tools.
In particular, the post-processing of the raw calibration data is discussed in detail, and a number of different
approaches are considered. The quadrant-based and full-field based methods described previously are
compared, along with a half-field method. In each case we examine a number of features, including the
trade off between ease of use (including the total number of measurements required) versus sensitivity /
potential signal to noise ratio. We also examine how some techniques are desensitized to specific types of
tool or mark aberration, and suggest how to combine these with non-desensitized methods to quickly
identify these anomalies.
There are two distinct applications of these tool calibration methods. Firstly, they can be used as part of the
tool build and qualification process, to provide absolute metrics of imaging quality. Secondly, they can be
of significant assistance in diagnosing tool or metrology issues or providing preventative maintenance
diagnostics, as (as shown previously) under normal operation the results show very high consistency, even
compared to aggressive overlay requirements.
Previous work assumed that the errors in calibration, from reticle creation through to the metrology itself,
would be Gaussian in nature; in this paper we challenge that assumption, and examine a specific scenario
that would lead to very non-Gaussian behavior. In the tool build / qualification application, most scenarios
lead to a systematic trend being superimposed over Gaussian-distributed measurements; these cases are
relatively simple to treat. However, in the tool diagnosis application, typical behavior will be very non-
Gaussian in nature, for example individual outlier measurements, or exhibiting bimodal or other probability
In such cases, we examine the effect that this has on the analysis, and show that such anomalous behaviors
can occur "under the radar" of analyses that assume Gaussian behavior. Perhaps more interestingly, the
detection / identification of non-Gaussian behavior (as opposed to the parameters of a best fit Gaussian
probability density function) can be a useful tool in quickly isolating specific metrology problems. We also
show that deviation of a single tool, relative to the tool fleet, is a more sensitive indicator of potential
The introduction of new techniques such as double patterning will reduce overlay process tolerance much faster than the
rate at which critical feature dimensions are shrinking. In order to control such processes measurements with
uncertainties under 0.4nm are desirable today and will become essential within the next few years. This very small error
budget leads to questions about the capability of the imaging technology used in overlay tools today and to evaluation of
potential replacement techniques. In this paper we will show that while imaging technology is in principle capable of
meeting this requirement, the real uncertainty in overlay within devices falls well short of the levels needed. A proper
comparison between techniques needs to focus on all of the possible sources of error, and especially those that cannot be
simply reduced by calibration or by repeating measurements. On that basis there are more significant problems than the
relative capability of different measurement techniques. We will discuss a method by which overlay within the device
area can be controlled to the required tolerance.
In a previous publication, we introduced Blossom, a multi-layer overlay mark (Ausschnitt, et al. 2006, ).
Through further testing carried out since that publication, Blossom has been shown to meet the requirements
on current design rules (Ausschnitt, et al. 2007, ), while giving some unique benefits. However, as future
design rules shrink, efforts must be made now to ensure the extensibility of the Blossom technology.
Previous work has shown that the precision component of Total Measurement Uncertainty (TMU) can be
reduced by using extra redundancy in the target design, to achieve performance beyond that of a conventional
box-in-box measurement. However, improvements that single contributor to TMU would not be sufficient for
future design rules; therefore we have also to consider the Tool Induced Shift (TIS) variability and tool to
tool matching contributions to TMU.
In this paper, we introduce a calibration artifact, based on the Blossom technology. The calibration artifact is
both compact, and produced by standard lithography process, so it can be placed in a production scribe line if
required, reducing the need for special sets of calibration wafers compared to other possible calibration
methodologies. Calibration is currently with respect to the exposure tool / process / mask, which is arguably
more pertinent to good yield, and less expensive, than calibration to an external standard; externally
calibrated artifacts would be straightforward to manufacture if needed.
By using this artifact, we can map out remaining optical distortions within an overlay tool, to a precision
significantly better than the operational tool precision, in a way that directly relates to overlay performance.
The effect of process-induced mark uncertainties on calibration can be reduced by performing measurements
on a large number of targets; by taking multiple measurements of each target we can also use the artifact to
evaluate the current levels of process induced mark uncertainty. The former result leads to an improvement
method for TIS and matching capability. We describe the artifact and its usage, and present results from a
group of operational overlay tools.
We show how the use of this information also provides further insight into the layout optimizations discussed
previously (Binns et al. 2006 ). It provides the current limits of measurement precision and mark fidelity
with respect to target redundancy, enabling us to use a predictive cost-benefit term in the optimization.
Finally, examining the bulk behaviour of a fleet of overlay tools, allows us to examine how future mark
layouts can also contribute to minimizing TMU rather than just precision.
We have demonstrated the feasibility of measuring overlay using small targets with an optical imaging tool has in
earlier papers. For 3&mgr;m or smaller targets, overlay shifts introduce asymmetry into the target image. The image
asymmetry is proportional to the overlay shift and so this effect can be used to measure the overlay.
We have used wafers built using production 45nm and 55nm processes to test these targets in production control
situations. Targets with different programmed offsets allow the necessary conversion between image asymmetry and
overlay shift to be determined empirically on the wafer under test. Measurements made using standard 25&mgr;m
bar-in-bar targets and 3&mgr;m in-chip targets agree to within 10nm (3&sgr;). By processing results from five or more fields
the agreement is improved to 5nm, a level which is limited by a mechanism other than random errors and which is
similar to differences between different styles of bar-in-bar targets.
Analysis of data from both in-chip and bar-in-bar targets shows similar patterns of overlay variation within the device
area. The pattern of overlay variation does not fit mathematical models of overlay as a function of location. The
total change of overlay within the field is 10nm, exceeds the overlay budget for critical layers at 45nm design rules.
This uncontrolled in-field variation in overlay must be reduced and ideally eliminated if process control is to be
achieved. A first step in controlling these errors is having an ability to measure them, and our data shows that this is
possible with targets no larger than 3&mgr;m in total size.
Improved overlay capability and sampling to control advanced lithography has accelerated the need for compact, multilayer/
mask/field/mark overlay metrology. The Blossom approach minimizes the size of the overlay marks associated
with each layer while maximizing the density of marks within the overlay metrology tool's field of view (FOV). Here
we describe our progress implementing this approach in 45nm manufacturing.
The feasibility of measuring overlay using small targets has been demonstrated in an earlier paper1. If the target is small ("smallness" being relative to the resolution of the imaging tool) then only the symmetry of its image changes with overlay offset. For our purposes the targets must be less than 5μm across, but ideally much smaller, so that they can be positioned within the active areas of real devices. These targets allow overlay variation to be tested in ways that are not possible using larger conventional target designs. In this paper we describe continued development of this technology.
In our previous experimental work the targets were limited to relatively large sizes (3x3μm) by the available process tools. In this paper we report experimental results from smaller targets (down to 1x1μm) fabricated using an e-beam writer.
We compare experimental results for the change of image asymmetry of these targets with overlay offset and with modeled simulations. The image of the targets depends on film properties and their design should be optimized to provide the maximum variation of image symmetry with overlay offset. Implementation of this technology on product wafers will be simplified by using an image model to optimize the target design for specific process layers. Our results show the necessary good agreement between experimental data and the model.
The determination of asymmetry from the images of targets as small as 1μm allows the measurement of overlay with total measurement uncertainty as low as 2nm.
A novel approach to overlay metrology, called Blossom, maximizes the number of layers measurable within a single optical field of view (FOV). As chip processing proceeds, each layer contributes a set of at least four marks, arranged symmetrically on concentric circles, to create a 90° rotationally invariant array of marks that "blossoms" to fill the FOV. Radial symmetry about the target center is maintained at each layer to minimize susceptibility to metrology lens aberrations. Overlay combinations among detectable marks within the target can be measured simultaneously. In the described embodiment, 28 distinct layers are represented within a 50μm square FOV. Thus, all the layers of a functional chip can be represented in a single target. Blossom achieves several benefits relative to overlay methods currently in practice:
* Compression (>30X) of area required for overlay targets. * Nullification of within-target proximity effects. * Suppression of optical mark fidelity (OMF) errors. * Reduction of sensitivity to across-target detection noise.* Elimination of overlay error random walk among layers. * Reference mark redundancy for detection flexibility and robustness. * Integration of multi-layer and within-layer overlay control schema. * Simplification of overlay recipe creation and management. * Capture and visualization of overlay performance through the entire chip fabrication.
Blossom results from 65-nm products in manufacturing are described.
A novel overlay target developed by IBM and Accent Optical Technologies, Blossom, allows simultaneous overlay measurements of multiple layers (currently, up to 28) with a single target. This is achieved by a rotationally symmetric arrangement of small (4 micron) targets in a 50 micron square area, described more fully in a separate paper. In this paper, we examine the lessons learned in developing and testing the Blossom design. We start by examining proximity effects; the spacing of adjacent targets means that both the precision-like Total Measurement Uncertainty (TMU) and accuracy of a measurement can be affected by proximity of features. We use a mixture of real and modelled data to illustrate this problem, and find that the layout of Blossom reduces the proximity-induced bias. However, we do find that in certain cases proximity effects can increase the TMU of a particular measurement. The solution is to ensure that parts of the target that interact detrimentally are maximally separated. We present a solution to this, viewing the problem as a constrained Travelling Salesman Problem. We have imposed some global constraints, for example printing front-end and back-end layers on separate targets, and consistency with the overlay measurement strategy. Initially, we assume that pairwise measurements are either critical or non-critical, and optimize the layout so that the critical layers are both not placed adjacent to any prior or intermediate-layer features. We then build upon this structure, to consider the effect of low-energy implants (that cannot be seen once processed) and site re-use possibilities. Beyond this, we also investigate the impact of more strategic optimizations, for example, tuning the number of features on each layer. In each case, we present on-product performance data achieved, and modelled data on some additional target variants / extreme cases.
Currently, overlay measurements are characterized by “recipe”, which defines both physical parameters such as focus, illumination et cetera, and also the software parameters such as algorithm to be used and regions of interest. Setting up these recipes requires both engineering time and wafer availability on an overlay tool, so reducing these requirements will result in higher tool productivity.
One of the significant challenges to automating this process is that the parameters are highly and complexly correlated. At the same time, a high level of traceability and transparency is required in the recipe creation process, so a technique that maintains its decisions in terms of well defined physical parameters is desirable. Running time should be short, given the system (automatic recipe creation) is being implemented to reduce overheads. Finally, a failure of the system to determine acceptable parameters should be obvious, so a certainty metric is also desirable. The complex, nonlinear interactions make solution by an expert system difficult at best, especially in the verification of the resulting decision network. The transparency requirements tend to preclude classical neural networks and similar techniques. Genetic algorithms and other “global minimization” techniques require too much computational power (given system footprint and cost requirements). A Bayesian network, however, provides a solution to these requirements. Such a network, with appropriate priors, can be used during recipe creation / optimization not just to select a good set of parameters, but also to guide the direction of search, by evaluating the network state while only incomplete information is available. As a Bayesian network maintains an estimate of the probability distribution of nodal values, a maximum-entropy approach can be utilized to obtain a working recipe in a minimum or near-minimum number of steps. In this paper we discuss the potential use of a Bayesian network in such a capacity, reducing the amount of engineering intervention. We discuss the benefits of this approach, especially improved repeatability and traceability of the learning process, and quantification of uncertainty in decisions made. We also consider the problems associated with this approach, especially in detailed construction of network topology, validation of the Bayesian network and the recipes it generates, and issues arising from the integration of a Bayesian network with a complex multithreaded application; these primarily relate to maintaining Bayesian network and system architecture integrity.
Pattern matching has long been a cornerstone of industrial inspection. For example, in order to obtain high accuracy, modern overlay metrology tool optics are optimized to ensure symmetry around the central axis. To obtain best performance, the metrology target should be as close as possible to that axis, hence a pattern recognition stage is usually used to verify target position before measurement. However most of the work performed to date has concentrated on situations where the imaging process could be described by simple ray-tracing, where the image is formed by albedo difference between surfaces rather than interference. However, current semiconductor technology requires optical identification of targets less than 30 microns (i.e. about 50 wavelengths) across, and of order 1 wavelength deep, and this description is no longer valid; interference and focusing effects become dominant. In this paper we examine these effects, and their impact on a number of different techniques. We compare image-based and CAD-derived models in the training of the pattern recognition system; CAD-derived models are of particular interest due to their use in “imageless” recipe creation techniques. Our chief metrics are precision and reliability. We show that for both types of pattern matching approach, submicron precision and high reliability is achievable even in very challenging optical environments. We show that, while generally inferior to image based models, that models derived from design data are more robust to changes caused by process variation, namely changes in illumination, contrast and focus.
Determining the focal position of an overlay target with respect to an objective lens is an important prerequisite of overlay metrology. At best, an out-of-focus image will provide less than optimal information for metrology; focal depth for a high-NA imaging system at the required magnification is of the order of 5 microns. In most cases poor focus will lead to poor measurement performance. In some cases, being out of focus will cause apparent contrast reversal and similar effects, due to optical wavelengths (i.e. about half a micron) being used; this can cause measurement failure on some algorithms. In the very worst case, being out of focus can cause pattern recognition to fail completely, leading to a missed measurement.
Previous systems to date have had one of two forms. In the first, a scan through focus is performed, selecting the optimal position using a direct, image-based focus metric, such as the high-frequency component of a Fourier transform. This always gives an optimal or near-optimal focus position, even under wide process variation, but can be time consuming, requiring a relatively large number of images to be captured for each site visited. It also requires the optimal position to be included in the range of the scan; if initial uncertainty is large, then the focus scan needs to be longer, taking even more time.
The second approach is to monitor some property which has a known relationship to focus. This is often calibrated with respect to a scan through focus. On subsequent measurements the output of this secondary system is taken as a focus position. This second system may be completely separate from the imaging system; the primary requirement is only that it is coupled to the imaging system. These systems are generally fast; only one measurement per site is required, and they are typically designed so that only limited image / signal processing is required. However, such techniques are less precise or accurate than performing a scan through focus, and they are also susceptible to effects caused by variations of the wafer under test, e.g. variations in stack depth.
A fast, precise system for measuring focus position, using the imaging optics, has been developed. This new system achieves better accuracy than previous indirect techniques, significantly faster than executing a scan through focus. Its output is linear with respect to focus position, and it has a very high dynamic range, providing a direct estimate of focal position even at large focus offset. It also has an advantage over indirect systems of being an integral part of the imaging system, eliminating calibration drift over extended periods. In this paper we discuss the mathematical background, optical arrangement and imaging algorithms. We present initial performance results, including data on repeatability and time taken to measure focus.
Overlay metrology is a very demanding image processing application; current applications are achieving dynamic precision of one hundredth of a pixel or better. As such it requires an accurate image acquisition system, with minimal distortions. Distortions can be physical (e.g. pixel size / shape) or electronic (e.g. clock skew) in nature. They can also affect the image shape, or the gray level intensity of individual pixels, the former causing severe problems to pattern recognition and measurement algorithms, the latter having an adverse effect primarily on the measurement itself.
This paper considers the artifacts that are present in a particular analogue camera, with a discussion on how these artifacts translate into a reduction of overlay metrology performance, in particular their effect on precision and tool induced shift (TIS). The observed effects include, but are not limited to, banding and interlacing.
This camera is then compared to two digital cameras. The first of these operates at the same frame rate as the analogue camera, and is found to have fewer distortions than the analogue camera. The second camera operates with a frame rate twice that of the other two. It is observed that this camera does not exhibit the distortions of the analogue camera, but instead has some very specific problems, particularly with regards to noise.
The quantitative data on the effect on precision and TIS under a wide variety of conditions, is presented. These show that while it is possible to achieve metrology-capable images using an analogue camera, it is preferable to use a digital camera, both from the perspective of overall system performance, and overall system complexity.
Network-centric architectures are defined by the complete absence of a traditional central data fusion site and also, in general, a central communication facility. Instead, the data fusion is performed at each network node and these nodes communicate on a strictly point-to-point basis. The network topology, which may be dynamic, is assumed to be unknown. These governing constraints imply a fault-tolerant, scalable, and modular system. However, such systems are prone to possible inconsistent fused estimates as a consequence of the well-known rumor propagation problem. The algorithmic challenge is to combat this problem without sacrificing the aforementioned benefits. This has led to the formulation of a technique known as Covariance Intersection (CI). Most recently, CI has been integrated with the Kalman filter to produce the Split CI algorithm - a general solution to decentralised data fusion in arbitrary communication networks. These algorithms have not yet been evaluated outside of a limited simulation environment. The purpose of this paper is to present a study of their relative performance in a hardware-based decentralised sensor network system.
The paper will describe a number of indoor experiments that involve tracking a ground target by means of multiple, networked, wall-mounted cameras. High precision ground truth target positions are available from a laser-tracking device. The experiments will evaluate Kalman, CI, and Split CI algorithm performance - measured in terms of consistency, convergence and accuracy - with respect to a range of static and dynamic network topologies.
In this paper we describe a simple physical test-bed, developed to allow practical experimentation in the use of Decentralised Data Function (DDF) in sensor-to-shooter applications. Running DDF over an ad hoc network of distributed sensors produces target location information. This is used to guide a Leica laser-tracker system to designate currently tracked targets. We outline how the system is robust to network and/or node failure. Moreover, we discuss the system properties that lead to it being completely “plug-and-play”, as, like the distributed sensor nodes, the “shooter” does not require knowledge of the overall network topology and can connect at any point.
We have been developing a decentralised architecture for data fusion for several years. In this architecture, sensing nodes, each with their own processing, are networked together. Previously, we have researched fully connected networks, tree-connected networks, and networks with loops, and have developed a range of theoretical and empirical results for dynamic networks. Here we report the results obtained from building and demonstrating a decentralised data fusion system in which the nodes are connected via an ad hoc network. Several vision based tracking nodes are linked via a wireless LAN. We use UDP to establish local routing tables within the network whenever a node joins, and TCP/IP to provide point to point communications within the network. We show that the resulting data fusion system is modular, scalable and fault tolerant. In particular, we demonstrate robustness to nodes joining and leaving the network, either by choice or as a result of link drop-out. In addition to experimental results from the project, we present some thoughts on how the technology could be applied to large scale, heterogeneous sensor networks.
Previously, we have developed techniques for Simultaneous Localization and Map Building based on the augmented state Kalman filter. Here we report the results of experiments conducted over multiple vehicles each equipped with a laser range finder for sensing the external environment, and a laser tracking system to provide highly accurate ground truth. The goal is simultaneously to build a map of an unknown environment and to use that map to navigate a vehicle that otherwise would have no way of knowing its location, and to distribute this process over several vehicles. We have constructed an on-line, distributed implementation to demonstrate the principle. In this paper we describe the system architecture, the nature of the experimental set up, and the results obtained. These are compared with the estimated ground truth. We show that distributed SLAM has a clear advantage in the sense that it offers a potential super-linear speed-up over single vehicle SLAM. In particular, we explore the time taken to achieve a given quality of map, and consider the repeatability and accuracy of the method. Finally, we discuss some practical implementation issues.
We have developed techniques for Simultaneous Localization and Map Building based on the augmented state Kalman filter, and demonstrated this in real time using laboratory robots. Here we report the results of experiments conducted out doors in an unstructured, unknown, representative environment, using a van equipped with a laser range finder for sensing the external environment, and GPS to provide an estimate of ground truth. The goal is simultaneously to build a map of an unknown environment and to use that map to navigate a vehicle that otherwise would have no way of knowing its location. In this paper we describe the system architecture, the nature of the experimental set up, and the results obtained. These are compared with the estimated ground truth. We show that SLAM is both feasible and useful in real environments. In particular, we explore its repeatability and accuracy, and discuss some practical implementation issues. Finally, we look at the way forward for a real implementation on ground and air vehicles operating in very demanding, harsh environments.
Technical details of laboratory based robotic system for researching decentralized Simultaneous Localization and Map building (SLAM) are provided. The main components of the system are Pioneer (ActivMedia) robots, laboratory environment for mapping, laser tracking system for testing the SLAM accuracy and a suite of SLAM software algorithms. The system is used to provide a demonstration and initial practical results of decentralized multiple-platform SLAM. The paper concludes that useful system has been set-up for researching this technology area. Further, the demonstration highlights important benefits of multiple- platform decentralized SLAM over a single platform approach. These include an increase in map accuracy, an improvement in the completeness and timeliness of the map, and an increase in platform accuracy although that platform was not extrinsically sensed. Future research areas are discussed.