Improving Yields and Tool Uptime with Relative Humidity Sensor in Semiconductor Environments
As presented during the 2020 SEMI Global Smart Manufacturing Conference. Ms. Vidya Vijay, Program Manager, CyberOptics, discusses the importance of measuring relative humidity to improve yields and increase tool uptime in semiconductor environments with the WaferSense® Auto Multi Sensor™.
VV: Thank you everyone, I’m Vidya Vijay, Product Manager at CyberOptics. At CyberOptics, we design and develop 3D high-precision sensors, and also inspection and metrology systems for SMT and Semiconductor markets. Today, we are going to talk about one of the main pillars of Smart Manufacturing. We’ll touch on the sensing part of it, but we’ll also talk a bit about how we had to work on the other two pillars as well, to help our customers. And we’ll see how we can increase yield and tool uptime by monitoring relative humidity.
In the wafer fabrication area, when the humidity level fluctuates, there are a host of problems that can occur. They are very sensitive to moisture in the wafer environment. Excess humidity can cause oxidation and corrosion, improper adhesion of photoresists, and problems during bonding. All of this accumulates to operational failure in the assembly. And as the nodes get smaller and smaller, it is more critical, much more critical, to characterize how much relative humidity is there. And also, to understand what the contaminates are. This includes all 29 nanometer fabs and below.
We are talking about a clean environment; we are talking about clean room and fab areas, and controlled humidity areas. But there are still sources of humidity, and let us look at that.
During photoresists, in mass jobs, the photoresists can absorb water vapor so rapidly due to their hydroscopic nature. It can absorb anything that it comes in contact with. In wafer fabrication areas, where there are spinners that spray on a wafer surface, that can result in water vapor across the entire surface of the wafer. In photolithography, photoresists can, when there is water vapor that is absorbed in the photoresist, have improper adhesion and that is not very helpful when the patterns you want on your wafers don’t stick to them. Also, as you know, when there is water vapor in your tool, pumping them down, getting them to high vacuum, can slow things down, which is your Cryopump that can become very slow. There is also your protective equipment and many other sources based on the type of processes that we can find some condensation or humidity observed.
So it is important to monitor this, because if we don’t know where the humidity is, there are many adverse effects. In many of the fabs there is a nitrogen purge, which is the standard medium for purging surfaces to clear the atmosphere from other gas or liquid contaminants. XCDA is extreme clean dry-air atmosphere. So in these atmospheres, there is humidity, there are many issues from the characterization. The resists characterization can change. Your bake-out times can change. Or the entire process can become more random and less predictable. Here you can see when there is some humidity or water vapor present on the wafer. If you further use this wafer to make ICES, there’s going to be many failures. Also, when the wafers are stored in the polymer cases, it can have a haze effect. As you keep storing the wafers in the clear case, you can have some light scatter, which increases over time, and this creates a time-dependent haze. This is common not just for wafers, but also in reticles. You can see how the haze looks like on a reticle, which is quartz, a glass material. This haze is not ideal for making wafers, and makes the entire process very random and less predictable.
This has been a problem for a while. It’s not a new problem. We know we have to monitor the humidity.
Challenges with existing RH solutions:
There are many ways of monitoring the humidity and characterizing when and what makes the humidity increase as well. There are some hand-held RH meters, which if you want to monitor in the FOUP, you have to open the FOUP. Opening it immediately can increase the humidity. Also, we want to monitor the humidity in the tool – across the tool, at different places. You will have to open up the tool. Not all the spots in the tool are easily reachable, as the tools are huge. Also, there are some home-grown fixtures where you have a test wafer, you have a probe and everything connected, but the limitation is that you are unable to monitor the humidity across the FOUP. You want to know how the wafer, when it sits on 3 or 6 or position 15, wherever it is, you want to be able to monitor it from top to bottom and across the wafer, and not just on one point. There are also solutions like where on the wafer surface you have RH recorders, but we want this data to be real-time, not one-time data. We want it to be real-time because we want to correlate the increase in humidity with the process that is happening.
There are other ways where you have the FOUP where the wafers are stocked. There are inlet ports where nitrogen or some other inert gas is purged into, in order to clean the FOUP of any particle or any other contaminants, and then there is an outlet port. You can have humidity monitored at the outlet port, but what happens when we have just a single point and again, we want to make sure if it is a 300mm wafer, 200, or 150, that you are able to monitor and know the humidity across the entire wafer and in FOUP from top to bottom. There are issues where you can see plating uniformity. Uniformity is also a big concern. We want to make sure humidity is not the culprit that causes other failures in the process.
Solution for measuring and monitoring RH:
We heard many of these problems from our application and customer base, and then we thought, we need to have a sensor where we have humidity sensors across many locations of the wafer. So we have a sensor in a wafer format, that can travel anywhere a silicon wafer will travel. We also have this in a reticle format, so it applies to reticle and tools. It can travel in any tool where a reticle will be stored or stocked, not just in the FOUP. Importantly, as it travels, it’s communicating in real-time. So if they are doing the maintenance purge and something is happening, we are able to see the data in real-time, which is very important because they need to know how much to purge and to characterize their maintenance cycle.
And while we are thinking about the humidity, we have these sensing modalities separated in the wafer form, but it helps for some applications if we also know the vibration. There is a vibration sensor and leveling sensor in the WaferSense Auto Multi Sensor (AMS) and the ReticleSense AMSRQ as well. So if we know the vibration as it is traveling across many chambers and stations in the tool, that may help as well. Most of our sensors are single modality sensors, but this one is a multi-mode sensor because it can measure humidity, vibration and leveling. We designed this based on the need. It was important for customers to know and characterize not just humidity, but also vibration and leveling. We will see more of this in later slides. As it goes, it records data and communicates wirelessly via Bluetooth, so the customer can see what is happening when they do a process change, or an improvement.
There are five humidity sensors, one in each of the levers and another one, a fifth in the center. The fifth one usually runs hotter than the other sensors because it is closer to electronics so we had to provide some type of correction curve for that sensor. In this example here, we had a case where the customer had to characterize their purge procedures, what they do at the maintenance cycle, and what they do at the process cycle. They had a maintenance purge, so as they started purging, you can see how the humidity dropped from 40% to 15%, even closer to 5% here. And it started increasing as soon as they opened the door. They opened the door to do an evaluation, and then they started the process purge. But the process purge, that procedure, got them only to around 20% relative humidity. So they stopped the purge, closed the door, and then started the procedure and brought the humidity down as they established a new cycle. And while all of this is happening, the graph is plotting as the AMS sensor is traveling through the tools. You can see one of the sensors is running at a higher humidity, and we give the correction curve, and that’s because it is much closer to the electronics. This is a good case where there is so much change, and each process might have extremely dry cases, and they might require they keep at a certain percentage, so this is helpful to characterize it.
Here you can see how it is important to note the leveling in both x and y, for uniformity and so on. Similarly for the vibration, we can see the profiles in x, y and z. We talked about how we need to measure humidity across the wafer, not just at one point. So that is very critical too.
We have these five sensors and real-time measurement because there was a need to identify where water vapor and humidity is coming from. That is very important. We can just measure the humidity, but since it is wireless and can go wherever the wafer goes, it gives them the correlation. The correlation is super important because they can actually rectify and prevent the problem ahead of time. There is also the vibration and the inclination functionalities in both the wafer and reticle formats.
We strived hard to keep it to the wafer size, and to the SEMI standards for weight so it can be used as close to how they would use a silicon wafer. Because they chuck it, and these are chuckable and are as thin as can be. It’s not exactly as thin as a silicon wafer itself, but we try to keep things to minimum thickness (different devices vary in thickness), but it is important to be able to go to places where a wafer or reticle would normally go.
You can see a couple of tools here. If we don’t have something wireless, these are all steps to smarter manufacturing. We need to be wireless. We want to be able to go into these big tools and understand where the problems are and help with environmental issues like humidity and temperature, and tell them what is happening as they go through the process. If it is the reticle sensor, it can go through the micro environment and can travel through the reticle library and it can go through anywhere in the reticle stocker, plus it can also sense inside a reticle pod, that is where reticles are stored. And in a SMIF pod, so it can measure inside a case. Basically, we go wherever the reticles and wafers go, and we want to be as close as the SEMI spec for these actual reticles and wafers.
Here we had measurements in this case. You can see how the opened the FOUP and they did a purge, but when you do a purge you would expect the humidity to go down. But there was an issue they were able to figure out with our real-time software, where you can drop these markers wherever so you know what is happening and there is a record of what is going on. There are also ways that you can compare versus last time you did it. You are measuring something real-time, but you also want to look at when was the last time it worked well. You can compare and see what is going on. Here you can see that the purge wasn’t uniform, so there are probably some liquid contaminants or particles present in the FOUP still.
We can do this similarly for the vibration application. For vibration, we can plot x, y and z and measure the g’s. Here we can see the graphs both before and after tuning. For example, you can see some vibration here, and a good 88% improvement after tuning. Also again, in the y direction, you can still see ~ 64% improvement. Z gives us the best improvement after tuning here ~98%. In the end, after tuning the robot all your vibration is reduced. Anytime vibration is reduced it means the wafer or the reticle is going faster, so it’s traveling less time. But it can also indicate other things not just the tuning – all of a sudden you are usually okay in the process with this amount of vibration, but there is huge amounts of vibration and then maybe it is the bearings or something else has gone bad. It is not just the tuning. There is probably more of a need, before things fall apart, like preventative maintenance can be done. Again, if the sensor can go wherever the wafer goes, it helps to pinpoint the location. And when you see signals in real-time, you can say, okay, it was going from the end effector to the next chamber, and so on. You can indicate where the vibration came from as well. And in this case, the increased yield and throughput, 203 wafers per hour per tool. This is huge savings.
We got the x, y, z and RMS (Root Mean Square) in g’s for the vibration, but it is important to give FFT curves, of frequency response, in the same amplitude in the frequency response. If there are a couple of different motors or something else, it can help them identify the source better. There is also the FFT which is computed in the software that is provided, so we can identify the source of the vibration, not just if the vibration is present.
The leveling is more like a bubble level, but extremely accurate to 0.003°. Leveling is important because you don’t always want to be level – some applications may want it to be tilted a bit. With this, you can calibrate how much of the tilt you want. Or if you need to be at perfect level, this is very helpful to tell you how level you are. It gives you a visual and also a plot to make sure how much is your x and y. And if something changes, this also helps if you need to re-tune your robot if it’s not placing it properly in a chamber, or if something else is happening or something else is loose. This can be an indication of other issues as well. If it is not level, a silicon wafer can get scratched, hit something, or cause particles if it is not placed level and there is a gap, and so on.
We saw this need and we talked about how important it is for us to sense environmental relative humidity (RH) and things like that. It is so important as the nodes get smaller, to know about particles and where the particles are coming from, plus the leveling, inclination and vibration to help increase yield and reduce down time. It is not a binary thing for vibration. You can have some vibration and still be okay. It may help to use sensors like AMS. These multi-sensors can help characterize and qualify equipment and robot transfers. It can reduce maintenance cycles. Because if we know this is the vibration that is expected, and now it has exceeded it, now a longer maintenance cycle has to be called in right away before you end up with more expenses because of a tool crash. It also reduces resource needs. Before this wireless technology, customers had to move and open up tools and when the majority of the tools run hotter, we are waiting for the tool to go down in temperature. We are waiting for the tool to pump down to be at atmospheric temperature. You still have to cool down for our sensors to be in the tool, but at least you don’t have to open up and re-clean then look for particles again and so on. All of this can happen with the tool intact. That is important. It helps the processes be streamlined. This again, is a step towards smart manufacturing. We want to be able to automate, and in order to automate, we want that real-time signal. It can either be humidity or vibration, so we can be proactive about issues before they happen, and that’s how we can save time and money. The tools are expensive and we want to be able to keep them running at all times without any yield issues. For example, an issue in a mask can get carried over to wafers and the wafer lots. It could be very disastrous.
We talked about sensing, and the second pillar is communicating; signal connecting. Right now our devices communicate Bluetooth, but in the future we are planning to be more Ethernet and IoT types of things, where we can do even more smarter applications and indications to the customer. We have this UI for different applications, but we also have a Software Development Kit. Because it is harder to customize it for different customers, and if we give them a library, and then they can see if the vibration exerts more than the set baseline, it would alert a couple of engineers, to let them know to go check or to call for an alarm for immediate maintenance. So all three pillars are very important, and thinking more in that way, we want to go more Ethernet and go towards that technology. If there are fab people and OEM’s listening, it is harder with security these days to go into Ethernet right away. Bluetooth is much more easy to penetrate because of security reasons, so there are other things not just about connectivity and being smart, security comes into a play. And these are the things we are thinking about right now at CyberOptics. How to be secure, how to give more intelligent feedback, and how to continue to improve our accuracy even further. As the nodes get smaller, our need to become even thinner is key. We want to be thinner and thinner, and we also want to be even more accurate. Our wireless sensors are the most efficient and effective available and quality of sensing is super important.
Apart from that, I would like to conclude with the fact that we are not only measure leveling, vibration and relative humidity, but we also have wireless sensors for particles, gapping, robot teaching and a resistance sensor for etching applications where you want uniformity. Thank you so much for the opportunity.