The second face-to-face General Assembly meeting of the ECSEL JU project iRel40 took place in Ljubljana, Slovenia, from Monday November 7^th to Thursday November 10^th 2022. It was organized jointly with a support agency by the Slovenian project partner Jožef Stefan Institute.  The meeting started with a Get Together networking event on Monday 7^th 2022 late afternoon at the meeting hotel in Ljubljana. Partners were able to start initial discussions in a pleasantly relaxed atmosphere (Figure 1).

Figure 1: First discussions during the networking event. In the following 3 days, technical topics were discussed and analyzed in detail (left and middle). The local organizers together with the iRel40 leadership team (right).

About 90 participants took part in the event in Ljubljana, which means that most partners of the consortium joined the project event and took the opportunity of physical togetherness and active discussion. Representatives from all 13 iRel40 project countries were present.

Especially the key technical experts were onsite (Figure 2). During the General Assembly, the status of the iRel40 project was summarized and new innovative ideas such as the importance of digital twins and artificial intelligence (AI)/machine learning (ML) algorithms in terms of their impact on reliability were shared and discussed.

Figure 2: A moment with everyone who contributed face-to-face to this very effective and efficient event.

iRel40 is a project that considers reliability along the value chain from wafer to chip to package/module to board/system/applications. The first meeting day started with the General Assembly in the morning, followed by an overview on the project status, the outcome of the 2^nd year review as well as the status of upcoming deliverables, milestones and reporting. Upcoming actions for the last year were presented and discussed. Especial attention was given to the status of exploitation and potential project impact. In addition, a summary has been presented by the leaders of the five running technical work packages, focusing on the main results of the last period. These include the results on Modelling and Simulation, Materials and Interfaces, Smart Production, Holistic Reliability Testing as well as the 34 Use Cases and their impact. These two first blocks were held hybrid, which allowed remote participants to align with the project status. The technical workshops that followed this first part were held in person to focus on technical exchange and to leverage discussions.

Eight technical workshops were organized during the meeting, covering the five ongoing technical work packages. These workshops, prepared by different iRel40 experts, were brought together by the different work packages WP2 to WP6 to encourage technical exchange. The workshop program started with the status of the use cases to have their progress all three days in focus. The eight workshops focusing on reliability improvements included the following topics:

• Use cases: Status accompanied by poster sessions (WP6)
• Digital twins and impact on reliability (WP2)
• Artificial intelligence (WP2)
• Physics of failure (WP2, WP3)
• Intelligent manufacturing (WP4)
• Importance of KPI for testing (WP5)
• Material simulation (WP3)
• Test vehicles in iRel40 (examples from WP3)

Below, we briefly summarize the outcomes of these eight technical workshops:

Use case workshop (WP6: Inigo Ozgoidi, Klaus Pressel)

Work package 6 includes research and development on the 34 use cases improving reliability. These include 16 application-driven use cases from three domains Digital Industry, Energy, and Transport. In addition, the consortium works on 18 so-called industrial pilots that focus on the improvement of reliability in production and testing. The status of all use cases was presented by short highlight presentations of 3-5 minutes. For the three use cases that could not be presented, a live video was prepared by the use case owner. This session was followed by a poster session, which included all 34 use cases. Three special technical posters of Ph.D. work were added.

An award was introduced for the best three use cases and their corresponding posters. For this purpose, the on-site project participants voted for the use case they liked best based on the information they received during the General Assembly. Votes from 1-5 (with 5 being the best) in the three categories of Innovation Level, Cooperation, and Technical Quality were possible. The three best use cases were awarded during the social event (see Figure 6 below).

Figure 3:  Partners presenting their research related to use cases during the WP6 meeting.

In general, the 34 use cases made good progress in terms of the five overall iRel40 objectives on reliability improvement.

Workshop on digital twins  (WP2: Mostafa Moonir Shawrav)

Mostafa Moonir Shawrav (task leader from ams OSRAM) organized a workshop on Digital Twin technology at the M30 General Assembly in Ljubljana. The focus of this workshop was to get an overview of the development of digital twins for electronic components and systems from a reliability point of view. The WP2 leader, Heiner Moeller (FhG ENAS), introduced the digital twin concept and shared the results of the simulation-based digital twin approach developed by FhG ENAS and Sensitec. Jose Luis la Vara (UCLM) showed the development of a reliability-oriented approach for digital twin. Jonas Bredberg (EDRM) presented a system model of the DC/DC converter to predict the thermal response of the SiC MOSFETs and reliability. Bart Vandevelde (imec) showed the physics-of-failure-based package simulation model to predict the stress in the silicon die as a function of environmental variations. In the last presentation, Ingrid Maus (IFAG) presented the simulation of power cycling on board and lifetime estimation via a meta-model web app. In this workshop, the partners also discussed the possibility of writing a chapter in the upcoming iRel40 book on reliability focusing on digital twins. Overall, the workshop was very successful with a lot of intriguing discussions from the audience. This workshop concluded on the second meeting day.

After some announcements from the project leadership team, Day 3 started with the workshop on artificial intelligence applied to reliability improvement.

Workshop on Artificial Intelligence Applied to Reliability Improvement (WP2, Rok Hribar)

The 2nd iRel40 Workshop on Artificial Intelligence was the first face-to-face workshop under Objective 2.2 (Develop data-driven models for reliability improvement) and was intended to stimulate a broader discussion on best practices in the application of AI to reliability. Given the plethora of AI techniques developed in the project, it is important to categorize these methods in a meaningful way in order to provide guidelines for the future use of AI in reliability. In the first part of the workshop, Rok Hribar (JSI) gave an overview of the AI approaches developed and implemented in Tasks 2.2.3 (AI algorithm and model development) and 2.2.4 (Implementation of data-driven models in use cases). A categorization of the use cases was presented that enables a link between the AI methodology and the characteristics of the use case. This categorization triggered a lively discussion on the general guidelines of applying AI in the area of reliability and on specific cases that also need to be considered. The applicability of organizing test vehicles in a similar way was discussed and how it can be used to quickly find relevant resources in iRel40 deliverables. The second part of the workshop was dedicated to the presentation of specific artificial intelligence methods developed within the project. In this part, Chenyang Lai from IUNET-POLIMI presented their work on fusing expert knowledge with machine learning in a talk entitled "Development of PHM method for Power Electronic Devices based on Physics Informed Neural Networks (PINN)".

Workshop on Physics of Failure PoF (WP2: Frank Altmann)

The objective of this iRel40 task is on modelling of relevant failure mechanisms. This includes

• Identification of manufacturing and operation-related reliability risks at all integration levels
• Investigation and deep understanding of under-laying failure mechanisms (thermo-mechanical, electro- & stress-migration, electronic, chemical, …)
• Exploration and improvement of related physics-based degradation models.

Five presentations were given to share the results of the iRel40 project. Maurizio Millesimo from project partner IUNET-University of Bologna gave a talk on “TCAD Modeling of the Dynamic VTH Hysteresis in p-GaN Gate HEMTs”. Raffaele Coppeta (ams OSRAM) and Werner Grogger (FELMI) presented results on “Degradation of oxide-confined GaAs-based VCSEL emitter”. Leiming Du from TU Delft showed “Thermal-cyclic properties of different kinds of solder materials”. Heiko Knoll (Sensitec) presented a talk on “Discussion of offset drift related failure models as input for lifetime simulation concepts of an SO16 current measurement device”. Xavier Perpinyà from CSIC presented use case results on “PoF in power devices under overcurrent testing/aging for induction cooking”.

A common understanding for all these presentations was that “Knowing what really happens in microstructure is key for reliable electronics”. Failure modes can be complex and could appear on the system level only. Therefore, stress condition on the system level is required to understand the failure mode. Stress testing was performed to initiate failure modes and device degradation. Several new failure modes have been identified and explored. PoF models could be improved e.g. for

• Enhanced cohesive zone-based modelling approaches for interface delamination and cracking
• Enhanced Finite-Element (FE) Modelling including the Garofalo creep model for solder fatigue
• New modeling approach to quantify the oxidation-related degradation in VCSELs
• Enhanced TCAD modeling of trapping-related mechanisms by nonlocal Tunneling Model approach for GaN HEMTs.

Workshop on Intelligent Manufacturing (WP4: Josef Moser, Sabine Allmayer, Silke Braun)

The focus of the workshop organized by WP4 was on sharing results on improving production and development with respect to reliability. During the workshop  the scorecard approach on how the project can impact the overall iRel40 objective “lower the failure rate ppm” was discussed. In addition, this approach can be used to get more insides which performance indicators are all affected by the performed project activities to improve reliability.

The impact of the solutions assessed in iRel40 is documented using a self-explanatory impact assessment form. The scorecard form reflects the performance indicators and main iRel40 KPIs in the following triangle to qualitatively assess the impact of each task:

- key development areas:

- multi-scale and multi-physics simulations for the physics of degradation

- AI-based control systems in advanced manufacturing

- materials, testing, and prediction

- smart sensing and Big Data Analytics

- prognostics and health management/ digital twin/ condition monitoring.

- Phases of bathtub curves (early life/ useful life/ wear and tear).

- Project Objectives

Initial in-depth discussions gave the WP4 team good indications for further optimization of the template. The first KPIs per objective/ KDA and bathtub curve were provided by partners, which further formed then the basis for the following deliverables.

A Ph.D. work at IFAT covers “Modeling early life failure probabilities using covariates”.  Specific Field failures did not correlate with data from electrical Test, PCM (Process Control Monitoring), or SPC,  but there in most cases, there is a  correlation (like a category, not yet as root cause based on all  available equipment data) a found with equipment data

• This is why a high focus is put on equipment data and their exploitation using AI for various predictive options
  • process stability/quality
  • equipment stability/maintenance
• Production under real stability conditions guarantees the designed, developed, and qualified reliability criteria
• No surprises for the customer and the production manager

In terms of virtual metrology, process know-how, and data are required for prediction. It has been highlighted that “Correctly predicted process results increase production stability and therefore reduce reliability relevant events“. Thus, Root Cause Analysis is crucial and PoF is required for understanding: One example is identifying important (combinations of) parameters for the purpose of root cause analysis („explainable AI“) using SHAP-values (SHapley Additive exPlanation).

Workshop on Holistic Testing (Importance of KPI on testing, WP5 Susan Zhao)

The last workshop on Day 3 dealt with WP5 on Holistic Testing. The workshop began with four task leaders presenting the latest status of their reports and deliverables. Following the presentations, a panel discussion was organized by the WP leader, which focused on the impact of holistic testing assessment on reliability. The audience actively participated by asking the WP leaders and related task leaders some thought-provoking questions:

• According to you, what are the important impacts of holistic testing assessment on reliability? Could you give an example?
• Can the contribution be measurable?
• Will our test vehicles contribute to “failure rate reduction”? Which part of the Bathtub curve is considered?
• How should we define KPIs (key performance indicators) for the holistic testing assessment?

Many examples were given by the leaders of test vehicles to illustrate the impact of holistic testing on reliability in some aspects: Reducing the failure rate in manufacturing and increasing the lifetime of new products for new applications by developing smart and intelligent tests, increasing testing efficiency to shorten the development time of new innovative technologies, introducing new test methods to evaluate the reliability of products for special and critical applications, etc. The challenges of measuring and quantifying the improvement have been discussed extensively and some ideas of managing certain critical cases have been generated.

Overall, the insights gained from the workshop will contribute to advancing holistic testing practices and strengthening reliability across the ECS value chain. They will also guide all the partners in WP5 to finalize their deliverables in a consistent and efficient way.

Figure 4: Lively discussions during work package meetings took place. Here the WP5 team is discussing major outcomes towards KPI’s.

The iRel40 General Assembly meeting concluded with WP3 related workshops on “Material Simulation and Compact Modelling” and a Workshop on “Test vehicles investigated in iRel40 to improve reliability”.

Workshop on Material Simulation and Compact Modelling (WP3: Thomas Krivec, Julia Zündel, Elke Kraker)

Material understanding is a major challenge to reach reliability targets. In this workshop, four presentations were shared. Heiko Knoll from Sensitec presented results on “Aspects of assembly material variations for a SO16 device with respect to FE simulation results”. EMC aging models and data from literature are helpful inputs also for specific package applications, but particular differences must also be considered (e.g. process flow). Karl Fendt from AT&S presented results on “first steps in compact modelling”. He introduced their work on a small single and daisy chain micro-via test coupon for:

• Process qualification and Quality monitoring.
• Can observing resistance variations in healthy microvias be explained by geometric manufacturing tolerances? The first results strengthen the original hypothesis.
• Derived resistance distribution as a basis for decision threshold ok / not ok via.

Based on the first results ideas for modifications of the test coupon design were developed. Thomas Krivec and Julia Zündel from AT&S presented results on the Characterization and Modelling of Laminates & Prepreg Materials. They introduced work on their virtual lab.

Finally, Ingrid Maus from Infineon shared a presentation on “Stress simulation in packages: advanced material modeling for polymer materials”. Ingrid started with an introduction to the IPC/ JEDEC-9301 guideline for thermomechanical simulation. With respect to linear viscoelasticity she showed among others that Linear Elastic (LE) material modeling shows severe limitations when passing in a simulation “through the Tg region” (Tg: glass temperature). In respect to aging, she demonstrated the following:  When exposed to high temperature over a longer period of time, “oxidized” layers of molding compound are forming. “Oxidized” layers can be included in the simulation when the respective material characterization is done. These “oxidized” layers may change the thermo-mechanical behavior of microelectronics packages.

Workshop on Test Vehicles that improve reliability in iRel40 (WP3 perspective)

Elke Kraker from MCL introduced Test Vehicles from a WP3 perspective. In iRel40 these test vehicles are bricks related to:

• Investigation of material on chip, package, board/system level to improve reliability
• Investigation of a material interface to improve reliability
• Investigation of new equipment (e.g. failure analysis equipment, test equipment)
• Investigation of new processes on chip, package, board/system
  level to improve reliability
• Investigation of innovative methods
  (e.g. digital twins, algorithms, simulation approaches)
• Investigation of innovative testing
  (e.g. combined reliability testing)

Initial discussions and collection of test vehicles showed that more than 200 of those bricks are investigated in iRel40 and more than 100 in WP3. A template was generated with teams from WP1, and WP6, which form the envelope of the iRel40 project as well as WP3 to better describe the work on the bricks in iRel40.  The template includes the test vehicle description, which use case does it support, what is the link to the 5 overall iRel40 project objectives as well as a link to the key development area, and finally the exploitation potential of the work. Based on this template a test vehicle catalogue will be worked out in iRel40 and shared with the partners. For some of the bricks, there might be no direct link to any of the use cases, but such research work will be collected as a technology learning vehicle.  

A highlight of the face-to-face (f2f) meeting was the social event on the evening of the 3^rd meeting day. This took place at the Ljubljana castle, one of the world-famous cultural heritage sites in Slovenia.

Figure 5: A view of the castle as well as the iRel40 team during dinner.

Besides a dinner where technical discussions proceeded, also the “use case and best poster award” winners were presented (see Figure 6).

The best use case was IP2, entitled “Application of Image Processing Techniques in combination with ML/deep learning methods to enable automated classification of defect density images of GaN devices during Production”, led by KAI and supported by IFAT and UCLM. The second award went to the “PAN European Quality pilot line”. The third award went to “Application of design limits based, predictive design approach for the design of a reliable embedded in laminate device.”   Figure 6 shows the award ceremony during the social event.   

Figure 6:Use case award ceremony. The best use case was IP-2, led by Anja Zernig from project partner KAI (left), together with the project leader Klaus Pressel (right), who handed over the awards, second best use case was IP-4 led by Germar Schneider from Infineon, Dresden (middle), the third best use case was IP-11 led by Thomas Krivec from AT&S (right).

In conclusion, a very fruitful and excellently organized iRel40 technical meeting took place in Ljubljana. The partners now have a clear perspective on where to move in the last year of the project.