Annika Franziska Wandesleben1*, Delphine Truffier-Boutry2*,
Varvara Brackmann1, Benjamin Lilienthal-Uhlig1, Manoj Jaysnkar3,
Stephan Beckx3, Ivan Madarevic3, Audde Demarest2, Bernd Hintze4,
Franck Hochschulz5, Yannick Le Tiec2, Alessio Spessot3,
and Fabrice Nemouchi2
1Fraunhofer IPMS CNT, Germany
2Université Grenoble Alpes, CEA-Leti, France
3imec, Belgium
4FMD, Germany
5Fraunhofer IMS, Germany
∗Equal contribution
Abstract
This paper compares the contamination monitoring of the three largest microelectronics research organizations in Europe, CEA-Leti, imec andFraunhofer. The aim is to align the semiconductor infrastructure of the three research institutes to accelerate the supply to European industry for disruptive chip processing. To offer advanced edge AI systems with novel non-volatile memory components, integration into state-of-the-art semiconductor fab-rication production flow must be validated. For this, the contamination monitoring is an essential aspect. Metallic impurities can have a major impacton expensive and complex microelectronic process flows. Knowing this, it isimportant to avoid contamination of process lines. In order to benefit from thecombined infrastructure, expertise and individual competences, the feasibilityof wafer loops needs to be investigated.
Through a technical comparison and a practical analysis of potentialcross-contaminations, the correlation of the contamination measurement. Feasibility of Wafer Exchange for European Edge AI Pilot Linesresults of the research institutes is investigated. The results demonstratethat the three institutes are able to analyse metallic contamination withcomparable Lower Limits of Detection (LLDs). This result sets the foun-dations for smooth and fast wafer exchange for current and future needs, potentially not only within research institutes as well as with industrialand foundry partners. The present work pays attention to both surfaceand bevel contamination. The latter requires very specific contamination collection which was also compared. Nevertheless, some challenges needto be addressed in the future to advance and accurate contamination monitoring.
Introduction
The aim is to align the semiconductor infrastructure of the three largest microelectronics research institutions in Europe, CEA-Leti, imec and Fraun-hofer, in order to accelerate supply to European industry for disruptive chip processing. Contamination monitoring is an essential aspect of this alignment. Metallic impurities can have a major impact on expensive andcomplex microelectronic process flows. Therefore, it is important to avoid contamination of the process lines. To benefit from the semiconductor infrastructure, expertise and individual skills, the feasibility of wafer loopsneeds to be investigated. Additionally, to offer advanced edge AI systemswith novel non-volatile memory components, integration into state-of-the-art semiconductor fabrication production flow must be validated. Metalliccontamination can have a major impact within the microelectronic processflow, whereby the different chemical elements have various effects. There-fore, contamination of the process lines must be avoided (Bigot, Danel,& Thevenin, 2005; Borde, Danel, Roche, Grouillet, & Veillerot, 2007).To simplify the future exchange of wafers in-between research institutesand between institutes and semiconductor fabs, it is necessary to find outmore about contamination monitoring and possible cross-contamination.For this purpose, a technical comparison and a practical analysis of thepossible cross-contaminations is carried out in order to furthermore inves-tigate the correlation of the contamination measurement results of the threeinstitutes.
![](https://njbing-lab.com/wp-content/uploads/2023/06/TABLE1-1024x424.png)
Technical Details and Comparison
The common techniques for contamination monitoring are TXRF and VPD-ICMPS. The three largest microelectronics research organizations in Europe, CEA-Leti, imec and Fraunhofer, are equipped with VPD-ICPMS while imec and CEA-Leti additionally use TXRF tools. The type of tool, its set up and qualification depend on the contamination management strategy developed in each clean room.
The capabilities of the individual institutes are summarised in the following Table 8.1
Comparison TXRF and VPD-ICPMS Equipment for Surface Analysis
TXRF is ideal for high throughput applications as the measurements arebased on the interaction of electron beams and silicon surfaces, without chemical manipulation. This technique allows to analyse fast enough both standardand noble elements in automatic mode with the possibility to localize thecontamination on wafer with the mapping option. However lower limits ofdetection (LLD) are quite high, from 1E+9 to 1E+11 at/cm2.
Concerning the VPD-ICPMS technique, it requires different chemicalsolutions for the collection of standard and noble elements, so campaignsneed to be planned and there is no local resolution of contaminants. However,the collection of all metallic contaminants in a small droplet of chemistryinduces significantly improved LLD values for all elements.
![](https://njbing-lab.com/wp-content/uploads/2023/06/%E5%BE%AE%E4%BF%A1%E5%9B%BE%E7%89%87_20230612024427-1024x273.png)
![](https://njbing-lab.com/wp-content/uploads/2023/06/%E5%BE%AE%E4%BF%A1%E5%9B%BE%E7%89%87_20230612024452-1024x303.png)
To compare metallic contamination results obtained by the different institutes, the first goal was to compare LLDs of each element of each institute andhow it is experimentally determined. Indeed, LLD is the lowest concentrationat which an element can be reliably detected and is a key point for thecontrol of the metallic contamination at very low level. Depending on the equipment, there are several ways to determine the LLD, and hence the needfor comparing the capabilities of each institute.
For TXRF, LLD values are nearly identical for each element, as shownin Figure 8.1. As this technique is based on physical principles and sinceboth institutes have the same equipment (Rigaku TXRF), capabilities of bothinstitutes are the same. All LLDs are between 5E+9 and 5E+10 at/cm2. OnlyCa and Ag are a little bit higher because Ca comes from the manual wafermanipulation and Ag results from a high background noise on the TXRFspectrum near 3 keV (Lα1 ray of Ag at 2.983 keV).
In case of VPD-ICPMS technique, the LLD results are not the sameacross the three institutes. This can be explained by the fact that the techniqueis based on chemical collection and each institute has its own specific systemwith different approaches to the analysis and calculation of LLDs, as shownin Table 8.2.
![](https://njbing-lab.com/wp-content/uploads/2023/06/TABLE.png)
Figure 8.2 shows that the VPD-ICPMS LLDs of each institute arebetween 1E+6 and 5E+9 at/cm2, more or less three decades lower than TXRFones.
Differences observed across LLDs of each institute are due to the different techniques used and the different environments. The collection system atCEA-Leti is not full automatic and technicians have to transfer a small container containing the chemical droplet from the VPD to the ICPMS. This container has to be manually cleaned between collection and all these manual steps contribute to the increased Na, Mg and Ca levels of contamination. However, these specific LLDs are still lower than 1E+10 at/cm2 and the seelements are usually not critical for the microelectronic device performances.For imec, high values of Ti and V seem to be due to specific detector settings that favours minimal peak interference for Ti and V. For other elements,all imec LLDs are lower as they use a fully automatic tool without manualsteps. Fraunhofer has a comparable system to CEA-Leti, but it is still in themethod development process and the current analyses are done externally onan automated system.
Overall, the VPD-ICPMS LLDs of each institute are very low and comparable to industry standards and thus are sufficient for the metalliccontamination control in the microelectronic environment. One other important parameter is the recovery rate that has to be more than 95 % foreach of the elements. As each institute use the same chemical solutionfor the collection step, recovery rates are nearly the same and are very good (>95 %).
VPD-ICPMS Analyses on Bevel
For several years, wafer bevel contamination has become a challenge in theindustry and it is therefore an increasing issue for R&D institutes. Actually,in order to increase device density on a wafer, individual chips need tobe placed closer to the edge of a wafer limiting the waste of surface. Inaddition, wafers are increasingly processed by physical contact at the bevel,so this particular part of the wafer will need to be precisely controlled in thefuture. The full bevel area can only be analysed by VPD-ICPMS on bare Siwafers. Effectively, TXRF analysis of the full bevel is impossible becausethis technique is too sensitive to the topography and cannot quantify themetallic contamination localized on the fall of the bevel. The collection ofcontaminants at the bevel is a key point and each institute had to develop aspecific system for the analysis. Thus, there are major technical variationsbetween the collection systems used by the three institutes for the analysis ofthe bevel.
The Figure 8.3 shows the different techniques used by each institute forVPD collection on the bevel and the resulting different analysis surface.Therefore differences are also expected for the results of the VPD bevelanalysis. Imec analyses the same area front side and back side 1 mm andthe bevel, CEA-Leti analyses 5 mm front side, bevel and 1 mm back side. In Fraunhofer institute, the area is not defined yet as the method is still under development. The monitoring of the bevel is another promising analytical technique and will be mandatory for the safe exchange of wafers, as with thiscontrol the probability of cross-contamination is further reduced.
![](https://njbing-lab.com/wp-content/uploads/2023/06/%E5%BE%AE%E4%BF%A1%E5%9B%BE%E7%89%87_20230612030710.png)
Comparison of the LLDs for VPD-ICPMS bevel are shown in Figure 8.4.It shows that the LLDs are higher than those of the VPD-ICPMS surfacesince they are in the range of 1E+8 and 1E+11 at/cm2. However, the valuesare quite similar and only Ti and V are noticeable again for imec due to theirspecific ICPMS detector setting.
Cross-Contamination Check-nvestigation
In the frame of the present study, one equipment of each institute wasselected for the control of the metallic contamination. Therefore, each institute chooses the tool that is regularly involved in the production memory flowand most critical in terms of contamination.
So called “witness wafers” were generated by each institute with theselected tool by handling bare Si wafers through the tool. In this way, thewafers are subjected to the specific tool contamination process. The anal-ysis of the backside delivers information about the contamination by thehandling system (chuck and robot). The analysis of the front side providesinformation about a possible contamination of the chamber. Afterwards, eachinstitute characterises the metallic contamination of the wafers with their owntechniques and finally, the analysis results are comprehensively evaluated
Example for the Comparison of the Institutes
For the practical comparison of the measurement, the results of the three research institutes for a tool from Imec are presented as an example. Thetool is a multi-module macro inspection, metrology and review tool for thefront side of 200 mm and 300 mm wafers and additionally for the back side and edge of 300 mm wafers. The tool supports the inspection of patternedand unpatterned wafers.
![](https://njbing-lab.com/wp-content/uploads/2023/06/8.5.png)
Figure 8.5 shows the comparison of TXRF measurement obtained byCEA-Leti and imec for the inspection tool. There is a high agreement betweenthe values, demonstrating the comparability of the measurement results. TheTi measured by imec is assumed to be a handling contamination during themeasurement. Nevertheless, the concentration is low.Figure 8.6 shows the comparison of the VPD-ICPMS data for the backside surface of wafers. For the VPD-ICPMS, the results show noticeabledifferences. On Figure 8.6, only detected element at concentrations higherthan the LLD are reported; i.e. if an element is not detected in one of institute,it is not mentioned in the graph. The first conclusion is that more elementsare detected by VPD-ICPMS due to the lower LLDs. All the concentrationsare lower than 1E+11 at/cm2and are in accordance with TXRF results.The second conclusion is that the three analysed wafers have not the samecontamination. If CEA-Leti and imec found Ga, Ge and Sb, Fraunhofer didnot detect these elements. Imec and Fraunhofer quantified Al, Fe, Ti and Wwhereas CEA-Leti did not find these elements. The analysed wafers are nottwins because the cross-contamination process do not allow to contaminateeach wafers at the same concentration. Moreover, some wafers were morehandled and shipped than other and these differences impact the metalliccontamination.
![](https://njbing-lab.com/wp-content/uploads/2023/06/8.6.png)
![](https://njbing-lab.com/wp-content/uploads/2023/06/8.7.png)
Figure 8.7 shows the results obtained on the bevel. Contamination levelson the bevel are higher than those measured on the surface. In this example,results obtained by CEA-Leti and imec are in agreement when the elementsare detected by both institutes. Concentrations measured by imec are almosthigher than those of CEA-Leti, probably due to the different influencingfactors. At first, collection techniques are different and the droplet scannedareas are not the same. Moreover, the bevel of each wafers was probablycontaminated by the handling and the shipping. That is why concentrations obtained on the bevel were always higher than those obtained on the surface.The study of the bevel is very challenging and these results show the metalliccontamination due to the process in the selected equipment, but also thosebrought by the handling and the shipping.
Conclusiion
This study confirms that the three different institutes are able to analysemetallic contamination either by TXRF or VPD-ICPMS with comparableLLDs. This result is very promising for the exchange of wafers in the future.TXRF, with higher LLDs, did not show metallic contamination above 1E+11at/cm2. On the other side, due to very low limits of detection, VPD-ICPMSallows to observe different concentrations obtained by the different institutes.Nevertheless, these concentrations are very low. The cross-contamination in atool do not allow to contaminate wafers at the same level. Hence in the future,in order to compare more reliably the capabilities of different institutes,an inter-laboratory test with intentionally standardised contaminated waferswould be necessary. Moreover, all the measurements were done on “witnesswafers” and not on product-wafers. In the future, it will be necessary todevelop techniques able to analyse the metallic contamination on real wafers during their flow. In this way, CEA-Leti has developed a new system allowingthe metallic contamination control of the bevel of product wafers. (Boulard,et al., 2022) (FR Patentnr. U.S. Patent No 20200203190 A1, 2020).
Although some additional improvement is required to create a smoothloop between the research institutes, this work makes wafer exchange flowmuch easier due to the first experiences and contribute to the strengthening of the collaboration in current and future projects. Moreover, the conclusionof this study broadens the capabilities in terms of tool, process and expertiseaccess for potential industrial partners. Thus, an important milestone has beenreached in aligning the three research institutes to offer advanced AI systemswith novel non-volatile memory components
Acknowledgements
This study was fully financed by TEMPO project. The TEMPO projecthas received funding from the Electronic Components and Systems forEuropean Leadership Joint Undertaking under grant agreement No 826655.This Joint Undertaking receives support from the European Union’s Hori-zon 2020 research and innovation program and Belgium, France, Germany,Switzerland, The Netherlands.