IMPROVE Final Summary

IMPROVE ENIAC 12005 Final Summary Implementing Manufacturing science solutions to increase equiPment pROductiVity and fab pErformance  Introduction To keep the competitiveness of European SC manufacturers in the face of worldwide competition, key challenges must be addressed: maintaining cost decrease per function, reducing cycle times and time to market, improving process reproducibility and equipment effectiveness while reducing the environmental impact of factories. Manufacturing Science is the key enabler that will allow overcoming these challenges. IMPROVE (Implementing Manufacturing science solutions to increase equiPment pROductiVity and fab pErformance) is a focused 42 month project that answers to the "advanced line operations" industrial project of the sub-programme SP8 "Equipment & Materials for Nanoelectronics" of the first ENIAC call. A consortium representing major European SC manufacturers worked to define the main challenges to be addressed and concluded that: 

With the decrease in device dimensions, control of the variability of each process step now demands the measurement of devices printed on each silicon wafer. In order to enable a rapid feedback on the process recipe adjustment, these measurements must be available immediately after the process. Standard metrology, performed on stand-alone measuring tools cannot provide the needed information in an acceptable timeframe and therefore cannot be applied on a wafer per wafer basis. The only solution for controlling a leading edge process is to introduce the concept of Virtual Metrology where the actual dimensions of the silicon devices are no longer measured, but are computed in real time from the equipment parameters thanks to statistical models.



Considering the huge amount of capital expenditure represented by the process tools in a leading edge semiconductor facility (2 to 10 Billions of $), it is absolutely critical to maximize the use of these assets, both in term of throughput ("speed of processing") and quality of result (supporting improvements in process stability and reproducibility) and in term of its overall efficiency, simply measured by the total number of hours the equipment is up and running within a certain normalized timeframe. To improve current availability and reliability of production tools, it is necessary to move from the present Advanced Process Control (APC) and Fault Detection and Classification (FDC) approaches which work in real time to control the tools behaviour, towards a predictive approach. One of the tasks of the project will be therefore to develop a Predictive Maintenance system to support this new approach.

IMPROVE was built aiming to enhance European semiconductor fabs efficiency by providing methods and tools to better control the process variability, reduce the cycle time and enhance the effectiveness of the production equipment. IMPROVE focuses on 3 major development axes:  The development of Virtual Metrology (VM) techniques allowing the control of the process at wafer level whilst suppressing standard metrology steps.

Page 1

IMPROVE Final Summary  

The development of Predictive Maintenance (PdM) techniques to improve the process tools reliability whilst optimizing the maintenance frequency and increasing the equipment uptime. The development of Adaptive Control Plan (ACP) concepts, suppressing unnecessary measurement steps whilst dynamically improving the control plan efficiency.

For these 3 topics, models are developed and assessed for different process steps and equipment platforms in different manufacturing lines leading to the development of generic solutions. 

IMPROVE Consortium

The objectives of IMPROVE could only be achieved with a consortium involving all types of specific expertise required: Semiconductor Manufacturers, Software Providers and Academics (32 Partners over 6 countries).  6 major European SC manufacturers: LFoundry (France), INTEL (Ireland), INFINEON (Germany, Austria), AMS (Austria), Numonyx (Italy) now Micron, ST (France, Italy)  2 Institutes: Fraunhofer IISB (Germany), CEA-LETI (France)  10 Solutions Providers: PDF Solutions, Probayes (France), Camline, ISYST , InReCon (Germany), LAM Research, Lexas Research (Ireland), Techno Fittings, LAM (Italy), Critical Manufacturing (Portugal)  12 Academic Labs: EMSE-CMP, GSCOP, LTM CNRS (France), Univ Augsburg, FAPS-Erlangen (Germany), DCU-Dublin (Ireland), UNIPV, UNIMI, UNIPD, CNR-IEIIT, CNR-IMM (Italy), FH-WN-Wiener Neustadt (Austria)

The main idea, building the consortium with many different semiconductor manufacturers, was to associate them in the same pre-competitive project in order to:  Develop solutions that can address a wider scope of work (multiple toolsets studied, greater range of solutions explored) in the same timeframe by distributing the development work between these companies;  Specify and develop systems that will answer globally to the needs of the semiconductor industry thanks to the various business models and type of facilities represented by these nanoelectronics components manufacturers. If we consider the IMPROVE consortium as a whole, a wide spectrum of competences are present (end users, universities and research centres, solution providers) and all aspects of the development of solutions (Data acquisition, modelling, assessment, implementation) are covered. From a modelling point of view the collaboration of industrial players with universities and laboratories with a high competency level of statistical and physical modelling led to the development of innovative approaches that could then be evaluated on actual data in the SC manufacturing lines.

 Technical Achievements and Exploitation Potential Although IMPROVE is a development project not designed to lead to commercial solutions or software packages, the project partners could highlight the benefit they got from the project results in their industrial strategies and business plans. The examples below show the exploitation potential for the different stakeholders of the program.

Page 2

IMPROVE Final Summary Common architectures and fab-wide framework A barrier for sustainable joint activities is often the variety of fab infrastructures at different partners. In IMPROVE six IC makers with nine manufacturing locations are working together with 29 research institutions, academia and solution providers. Thus, substantial efforts have been taken to gain common technical ground and to ensure a long-term applicability of technical achievements:  In-depth technical surveys were carried out at each IC fab, covering the current status and the specific requirements for new APC technologies.  Surveys were complemented by collecting the state-of-the-art beyond the IC manufacturing area, e.g. in the electronics and automotive industry.  Based on survey results and enhanced by the specific collection of user requirements for virtual metrology, predictive maintenance and adaptive control planning, common architectures and specifications were developed and transferred to the technical work packages. These efforts, mainly driven by academia and institutes (Fraunhofer IISB), culminated in a fab-wide communication framework to support the research of reusable APC components. The framework was realized by the IMPROVE solution providers or instanced in their existing ones (Camline, PDF Solutions, Critical Manufacturing) and successfully tested with industrial partners.

Architecture for a fab-wide framework in IMPROVE Virtual metrology The collaboration between industrials, academics and institutes has been organized by defining "clusters" of partners, organized around the SC manufacturers. These clusters investigated novel statistical theories and their application on actual Fab data in form of feasibility prototypes. Some of these clusters are listed below: 1. INFINEON Austria, Uni Pavia, Uni Padova 2. INTEL, LAM Research, DCU 3. STMicroelectronics, PROBAYES, EMSE-CMP GC, GSCOP 4. MICRON, Uni Pavia, LAM Italy, CNR IMM 5. INFINEON Germany, Fraunhofer IISB 6. AMS, Fraunhofer IISB, EMSE-CMP GC The complementary expertise of the partners allowed analysis of anonymized fab data from the SC Manufacturers using classical statistical approaches and, also, more advanced methods and tools. Thanks to the IMPROVE project, it has been possible for the first time to test an exhaustive panel of statistics for data selection and process step modelling on real process data.

Page 3

IMPROVE Final Summary Also, the sharing out of the work allowed addressing a wide range of process steps (CVD, PECVD, and Plasma etching for various process operations...) and led to applicable modelling solutions, defining the best tools for each type of problem that was studied.

2 lots ahead etch rate estimation (Gradient Boosted Tree Model) at INTEL IMPROVE developments led to the actual implementation of prototypes in the different fab lines of the partners and provided evidence of the expected benefits in different domains: 

Real-time VM provides information on all processed wafers and its integration in the FDC (Fault Detection and Classification) system enables wafer to wafer control loops resulting in a better process control and stability (INFINEON, INTEL, STMICROELECTRONICS).



VM predicted values can replace actual metrology measurements and therefore allow reducing the number of standard metrology steps. In some feasibility tests (INFINEON - CVD), this reduction should be around 30%.



VM modelling activity also resulted in a better knowledge of the equipment parameters influencing the process under examination. In some cases, "hidden correlations" could be found which enable process quality improvement. The deployment of VM to other processes will result in a proliferation of such learning (INTEL etch). In addition, this in-depth knowledge permits correcting lifetime equipment variation by adjusting real time process variables (MICRON etch).

The exploitation plan will be based on the deployment of VM methods and their subsequent extension to other processes than those studied in IMPROVE. These further steps will also require the integration of the VM systems in the decision systems of the manufacturing lines as shown below.

Real Time Virtual Metrology Workflow - Functionality Value of VM Activities (INTEL)

Page 4

IMPROVE Final Summary Predictive Equipment Behaviour Like for the Virtual Metrology activity, the collaboration between industrials, academics and institutes in this area was also organized in "clusters". In this case, the collaboration between SC Manufacturers and academics has been focused on the evaluation of advanced statistical methods to predict equipment behaviour, based on the equipment parameters and history. In addition to the type of models developed in Virtual Metrology, it was necessary in this case to take into account non numeric parameters, like the equipment history. In the semiconductor arena, such a concerted attempt to model the equipment behaviour, putting together the complementary skills of a big consortium of industrials, institutes and academics was a world first. Therefore, a lot of innovative solutions have been developed and assessed using the SC manufacturers' data.

Real evolution of temperature difference (blue solid) and estimated evolution (black dotted) with probability distribution (green curves)- IFAT

Scheme of the Predictive Maintenance Module (Epitaxy IFAT)

Algorithms for the computation of an Equipment Health Factor and prediction / anticipation of failures could be developed and assessed for lithography, implantation, Epitaxy, etch and thermal treatment. Following Predictive Equipment Behaviour studies, many prototypes have been implemented in the partners' fabs. Evaluation of potential benefits have been made thanks to off-line simulation or real use in the production flow using the IMPROVE framework. The following points have been demonstrated:  Thanks to more effective preventive maintenance planning, based on equipment parameters it has been measured a potential increase of equipment availability in the range of 1%-5% (INFINEON, AMS, MICRON, STMICROELECTRONICS).  The computation of the Equipment Health Factor (EHF) and the capability to predict equipment failure also improve the cycle time by reducing unexpected failure. To gain all the possible benefits, the PdM system has to be integrated to the lot scheduling one. This next step is under investigation (INFINEON, STMICROELECTRONICS)  As positive side effect, the knowledge of equipment behaviour from the modelling work provided a better diagnosis capability and therefore an additional improvement of the equipment availability in the range of 1% (INFINEON)

Page 5

IMPROVE Final Summary

Visualization Tool and list of chambers with high Maintenance Probability Exploitation will be based on the deployment of the PdM approach, with a strong link with the Fabs' Equipment management system and lot dispatching system. Complementing the above activity, the Dynamic Risk Assessment and Control Plan developments provide an efficient way to improve the control of the level of risk while simultaneously optimizing the control sampling plan. Pilot implementation showed drastic improvement in cycle time (STMICROELECTRONICS, INFINEON) or increase of information quality (INFINEON).

ST “Wipper screen”, allowing the operator to choose the best lots to process. Defectivity cycle time was divided by 2, improving the level of control on the risk. Conclusion Thanks to the wide range of competencies found in the IMPROVE consortium, the project addressed all the different technical aspects (sensors, data acquisition, modelling) needed to introduce the concept of Virtual Metrology and Predictive Equipment Behaviour into SC manufacturing lines. The intensive collaboration on the modelling activity between SC manufacturers and Academics enabled the discovery of the appropriate algorithms for the proposed case studies. Based on this work, prototypes were built by Solution Providers. These prototypes and algorithms have been assessed on real fab data and show that future operational solutions are realisable. The top objectives of the project, related to the improvement of the Equipment effectiveness and the Process Control efficiency were demonstrated on pilot implementation in the partners manufacturing lines.

Page 6

IMPROVE Final Summary

 Dissemination During the course of the activities related to Virtual Metrology and Predictive Equipment behaviour the high quality and innovation level of the work was attested by over 90 contributions in the form of papers, presentations or posters in journals, conferences like AEC/APC (Advanced Equipment Control / Advanced Process Control), CASE (IEEE Conference on Automation Science and Engineering), IFAC (International Federation of Automatic Control), SEMI ASMC (Advanced Semiconductor Manufacturing Conference), IESM (International Conference on Industrial Engineering and Systems Management), Future Fab International Magazine. Three of these communications have been recognized by an award. This is evidence of the flourishing research activity in these topics. On the Fab-wide Framework, infrastructures and Adaptive Control Plan, and other subjects addressed in the project, thanks to the new vision brought by academics and the Fraunhofer, and their close and fruitful collaboration with industry players, IMPROVE also produced innovative results which have been widely recognized: the R&D work on this topic was reported in more than 20 publications and communications in different newspapers and conferences To involve Equipment manufacturers not members of the consortium, an Equipment forum has been started within IMPROVE (Fraunhofer IISB). Its activity led to a final technical workshop in Grenoble in May 2012, in collaboration with the FP7equipment oriented project SEAL, gathering more than 40 participants, with more than half-of-them being not "IMPROVErs". Discussion also took place with the ENIAC project EEMI 450 & EEM450 PR for activities on APC and Virtual Metrology.

 Anticipated Market Impact & Specific Impacts on European Economy Semiconductor Manufacturing in general and in Europe in particular faces stiff competition in a very demanding and dynamic market environment. The cooperation of IMPROVE partners has led to the solutions developed being tested in each of the Semiconductor Manufacturer facilities. Semiconductor manufacturers being the main beneficiaries of the output of the IMPROVE technology and solutions will be able to compete more effectively and efficiently with competitors, especially from the Far East and other geographies. On the other hand, those Manufacturers will need Solutions Providers to implement such high end Software and Hardware solutions and to use their high-tech services experience and knowledge. For these solution providers, the opportunity to propose leading edge systems developed in close cooperation with actual SC manufacturers will help in their gaining market share and secure their market position. As all partners in the IMPROVE project came from European companies and institutes, a positive impact in the European economy is expected from the deployment of the project outcomes. European Semiconductors Manufacturers, Solution Providers and Academics will be more agile, better informed and prepared to resolve the difficulties and shortcomings of existing manufacturing solutions for the semiconductor market. On top of these tangible benefits, for all members of the consortium, IMPROVE was a great opportunity to have a close cooperation with several partners with distinct know-how but similar objectives, bringing together new and interesting ideas, developing trust and fostering a sense of community. At the same time, it opened the door to the possibility of future partnerships.

Page 7