Controlling the margins in 300mm manufacturing
02/01/2004
Semiconductor companies converting to 300mm manufacturing will need to implement advanced process control to keep capital investment and operating costs low while maximizing the value of each wafer.
Semiconductor companies are wrestling with the economic challenges of converting to 300mm manufacturing. The economy of scale makes this conversion ultimately beneficial, but there is inherent risk. The large capital and operating investment required to build or retool a fab for the new diameter requires a high rate of return on the investment in next-generation tools. The value of each wafer could be more than double the value of 200mm wafers. This demands a level of process-control improvement proportionate to the increase in the intrinsic value of the processed wafers and the cost of manufacturing. The key will be to keep these costs low while maximizing the value of each wafer. Advanced process control (APC) will become the crucial manufacturing technology that supports these efforts.
AMD has pioneered the deployment of APC solutions as a part of its Automated Precision Manufacturing (APM) capabilities, an integrated suite of more than 200 patented or patent-pending technologies. State-of-the-art control technology is implemented throughout Fab 25, in Austin, TX, and Fab 30, in Dresden, Germany. The transition to 300mm manufacturing will provide the opportunity to develop next-generation technology that capitalizes on the lessons learned from the development of these applications. We will explore three areas of opportunity for control evolution: wafer-to-wafer control, fab-wide control, and the specific process design for APC. Changes to the production infrastructure required to support this new technology will also be presented.
Wafer-to-wafer control
The current state of APC within the industry has effectively dealt with lot-to-lot variation in many semiconductor-manufacturing processes. Such control ensures that each discrete operation produces material within a given specification window. The assumption of control at this level of granularity is that such results will lead to salable product at the end of line. For effective 300mm manufacturing, however, APC technology must evolve to optimize the performance of each process and mitigate all sources of systemic variation.
AMD's APM technologies have allowed for the rapid proliferation of lot-to-lot adaptive control, effectively eliminating systemic variation from the lot-mean controlled process variable(s). A degree of within-wafer control has also been achieved in some control implementations. Within-wafer control has been limited to processes that produce or receive a fairly consistent nonuniformity distribution from wafer to wafer and possess a means to compensate for process variables. These results have significantly increased the capability of the controlled processes. Run-to-run control, however, is limited in its capability to control wafer-to-wafer effects, both in terms of wafer-average quality differences and nonuniformity distributions that change from wafer to wafer.
Wafer-to-wafer control is the required technology to address variability beyond the lot level. It drills down into the process and minimizes error in its control objective by modifying the process recipe at wafer level. The unique process recipe prescribed for each wafer seeks to compensate for one or more sources of variation. One source is upstream disturbances to the quality of a given wafer, which may produce a shift in the mean or distribution of a variable that affects process performance. An example of this disturbance is variability in a deposition process that affects the post-etch thickness of a deposited film. Another source stems from a shift in the performance of the process itself. Wafer-to-wafer control compensates for these disturbances at the wafer level by modeling and predicting their impact on wafer quality, and compensating in advance for their effect by wafer-level recipe modification. This significantly reduces the overall distribution of post-process error, as depicted in Fig. 1.
 Figure 1. Reduction in post-process error with increasing control granularity. |
300mm process tools need to be designed with the capability to perform wafer-to-wafer control in mind. Its implementation requires the ability to change the process recipe for each wafer in an automated fashion. This is analogous to the ability to make automated recipe adjustments between the processing of wafer lots developed for run-to-run control. In the case of pure feedforward control of upstream disturbances, as well as compensation for known wafer-to-wafer process effects, the recipe for each wafer could be specified prior to the processing of a given wafer lot. To implement full wafer-to-wafer control including feedback, however, the ability to change the recipe for one wafer during the processing of another will need to be developed.
Wafer-to-wafer control also depends on increased visibility into the condition of the process under control, including the tool and wafer states. Certain control opportunities may leverage integrated-sensor or other process-tool data to measure and/or infer these states during processing. In cases where a source of variation requires direct measurement of a wafer, wafer-to-wafer control would require the sampling of each wafer processed through a given tool.
Integrated metrology is an attractive solution to support such measurements, because it is largely performed during the concurrent operation of the process tool and can measure each wafer without significant impact on throughput. In addition, timely metrology on the first wafers out of a process allows modification of subsequent wafers' recipes to compensate for tool disturbances. In either case, effective wafer-to-wafer control will depend upon the successful development and implementation of embedded metrology technologies.
Information systems integrated into the production environment also need to be updated to handle the volume of information associated with wafer-to-wafer control. Data generated from embedded metrology must be communicated to control applications in a timely manner to support wafer-level control calculations. Recipe settings recommended by these applications need to be passed back to the tools so their recipes can be modified on a wafer-by-wafer basis. Pathways between tools and manufacturing systems already perform this communication, but must be augmented to seamlessly interface embedded metrology devices, process tools, and external systems on a wafer-by-wafer basis. It is believed that the significantly improved control capability provided by wafer-to-wafer control will outweigh the considerable investment required for many critical operations.
Fab-wide control
The use of a scalable, fab-wide control systems infrastructure has already demonstrated over 10× the return on the initial investment, but much of the return to date has come from the control of individual processes. An increasing benefit will come from the integration of all fab processes into a cohesive manufacturing operation that maximizes product quality and performance while lowering overall production costs.
Manufacturing controls are most often implemented within the scope of a specific process. The proliferation of control in our fabs began with the deployment of single-process control applications with post-process measurements as their control objective. Once control spanned enough processes, a higher level could be realized by cooperatively combining two or more operations within the scope of a single control application. Feedforward and feedback control between operations allowed tighter adjustment levels at downstream operations while loosening control requirements at upstream processes. With APM, AMD seeks to evolve the benefits of such cooperative control, encompassing the fab as a whole, a concept referred to as fab-wide control.
 Figure 2. Components of fab-wide control. |
Fab-wide control, as shown in Fig. 2, encompasses many process-control applications, as well as cooperative technologies that enhance control performance and seek to optimize total operations. The evolution to the fab level will ultimately move beyond the control of inline process characteristics to the point of controlling electrical test and other end-of-line parameters. This ability to control the electrical properties of fabricated devices, electrical parameter control (EPC), represents the highest level of control possible. Implementation of EPC begins with the development of models that relate electrical test parameters to process performance as characterized by inline metrology. Given such a model, the EPC application seeks to optimize electrical test parameters by changing the targets of the inline processes, a method known in control theory as supervisory control. The electrical test parameters are effectively controlled to specified targets through the optimization of the inline targets, which are in turn controlled by the individual process controllers. With this relation between individual process control and electrical test-parameter control, EPC effectively controls from the fab level down to the wafer level.
AMD is researching a wide range of production control methods to optimize fab throughput. Metrology optimization with regard to process control is one avenue of research, since visibility into the performance of each process is required to perform effective process control. Dynamic adaptive sampling (DAS) is one approach to optimizing metrology capacity in support of control applications. This technology varies the sampling rate associated with each process in response to the changing conditions within the manufacturing environment. It ensures that each application receives adequate metrology to support its needs without unduly taxing metrology capacity.
A companion technology to DAS under development is dynamic production scheduling (DPS), which optimizes the movement of material through the process tools in support of process control and higher-level production objectives. DPS incorporates the process-control requirements of particular operations and wafer lots, the capability of each process tool, and the priority of the material in the line to optimize material movement. Such a capability would allow AMD to manufacture material with specific device properties at desired times, allowing for optimal response to the market demands on each product.
The implementation of fab-wide control will leverage the infrastructure built to support APC, but will require coupling to several other manufacturing systems to execute its control. Electrical test data needs to be correlated to inline metrology results to support EPC modeling. Cooperation between existing data-storage systems, or the creation of a new storage mechanism, is needed for the merging of these data sets. Recipe-management systems will also need to be integrated with EPC to manage the specification of process targets. Yield-management systems also should be operatively coupled to EPC to provide better insight into its calculations and results. DAS and DPS require communication to production sampling and dispatch systems, respectively. Though this all represents a significant investment in manufacturing systems' design and development, their harmonious operation should reach unprecedented levels of manufacturing optimization.
Process design for APC
Before the advent of APC, the overriding goal of manufacturing was to develop processes that were insensitive to as many variables as possible. Small perturbations in conditions would not lead to large changes in product quality, allowing a tool to process for as long as possible between recipe changes. Given the utility and availability of APC, the need to develop such robust processes was mitigated to a certain degree by a controller's ability to compensate for changes in process or wafer conditions. For next-generation processes, this trend will continue to the extent of designing processes that are sensitive to a prescribed set of recipe parameters for the express purpose of implementing APC.
Incorporating APC is predicated upon the ability to affect the output of a process by manipulating one or more of its recipe settings. Most processes have established recipe settings, or "control knobs," that are used to control outputs; run-to-run control has primarily used these existing variables. Many processes, however, do not currently have knobs to control wafer nonuniformity. Such effects may have represented a small part of the error budget on smaller wafers, but wafer nonuniformity will likely be a large component of variation at 300mm. To control wafer nonuniformity, some processes or process tools need to be modified to manipulate recipes and affect process uniformity.
 Figure 3. Schematic of AMD's gate CD control application. |
One example of a recipe setting that specifically supports APC is the gate critical-dimension (CD) control application developed at AMD. As indicated in Fig. 3, the etching process associated with this controller was modified to add an isotropic etch step that affects the resist CD. The time associated with this step became a recipe setting that has direct impact on post-etch CD, allowing for feedback CD control at the etching step. In addition, coupling control at the etching operation to that at the masking operation facilitated feedforward process control. This not only resulted in an even higher level of control but also relaxed the control requirements at the masking operation, given the ability to compensate for masking errors at the etching operation.
Advances such as the gate CD control application bring about the opportunity to leverage APC when the process is designed to take advantage of it. Such process design is even more important for advanced control methods. Wafer-to-wafer control will need to use control knobs at the wafer level to control wafer-to-wafer and within-wafer effects. Fab-wide control will need to make changes to many production decisions, ranging from process targets to lot dispatch. With these abilities to maximize control technology within an Automated Precision Manufacturing environment, AMD's APC technology will be evolved to meet and even exceed the challenges posed by 300mm semiconductor manufacturing.
Conclusion
The control of processes as isolated steps in a manufacturing flow is not adequate for 300mm processing; rather, cooperative control of all operations to achieve yield- and electrical parameter-based control objectives will become the norm. The implementation of such control technology represents significant control evolution and will ensure manufacturing of optimal products at lower prices. Given the continued success of APC technology at AMD, its adoption in 300mm manufacturing will allow for the rapid adoption and proliferation of APC technology in our next generation of Automated Precision Manufacturing (APM 3.0).
References
1. T.J. Sonderman, M.L. Miller, C.A. Bode, "APC as a Competitive Manufacturing Technology: Getting it Right for 300mm," Future Fab International, Issue 12, March 2002.
2. T.J. Sonderman, C.A. Bode, "Advanced Process Control: What's Next on the AMD Roadmap?" in C.J. Progler and B. Singh, editors, Advanced Microelectronic Manufacturing, SPIE, 2003.
3. T.J. Sonderman, C.A. Bode, "Advanced Process Control Technology Evolution Requirements for 300mm Manufacturing," AEC/APC Symposium Asia, 2003.
Christopher Bode is a member of the technical staff of Advanced Process Control at Advanced Micro Devices, 5204 E. Ben White Blvd., MS 608, Austin, TX 78741.
Thomas J. Sonderman is director of automated precision manufacturing at Advanced Micro Devices.