Long-term partnership propels multiple improvements in SiN deposition

Plasma-Therm’s large-area plasma-enhanced CVD tool is a parallel plate design utilizing a 13.56 MHz power source to generate the plasma.  Wafer temperature can be controlled from 100°C to 350°C

Assistance to RFMD by Plasma-Therm through its Customer Improvement Program (CIP) did not stop with the delivery of the tools. Although the process worked as specified, there were intermittent issues, primarily handling, which Plasma Therm helped to investigate and address. Each issue was eventually resolved (see table 1), and one of the outcomes of this investigation was for RFMD to establish a ‘standard’ process used for production that differs from the high rate process used for testing and qualification.

Upgrading the indexer

One of the issues that required resolution relates to the indexer. This integral part of the processing chamber is a rotating fixture that is loaded with wafers. After one is added, the indexer subsequently lifts and rotates to the next position and the next wafer is loaded.  Once completely populated, the load lock is isolated by a slot valve and the chamber is pumped and processing of all the wafers is initiated.

There were lingering, intermittent problem associated with the indexer, and both companies tried to come to a plausible, workable solution. Returning to the original design yielded a better overall replacement, but several intermittent indexing problems remained: Losing home position; indexing to the same placement position; and wafer breakage, due to misplacement because of loss of home position.  A further upgrade improved overall reliability, but intermittent faults persisted; primarily misplacement of wafers loaded onto the indexer was caused by inaccurate indexer positioning.

Engineers at both firms strived to get to the bottom of this, and through continuous investigations and testing, they were able to implement a final, very robust solution.  This involved installing a drive and higher resolution encoding system at RFMD.  Initially three systems were retrofitted in exactly the same way.  When fully loaded the indexer was seen to have a minor issue with increased load (weight), which affected the sensitivity of the encoding mechanism.  The approach taken in this effort was to normalize the load across all three tools and set the corresponding sensitivity for optimal operating conditions under load and teach each load position (see Figure 1). After several months of running with zero faults on these first three systems, RFMD’s engineers upgraded their remaining tools with the drive and encoding package.



Figure 1. A robot is used to load the wafers into the tool. Each position on the indexer is ‘taught’ and the positions stored in memory for consistent loading and unloading

Improving the indexer led to an increase in wafer throughput. Addressing this issue propelled average wafer uptime from 91 percent to more than 97 percent, which equates to a productivity increase of over 10,000 wafers per year per tool (see Figure 2).

Figure 2. A small gain in uptime can translate into a throughput increase of thousands of wafers per year

Showerhead improvements

Working together, Plasma Therm and RFMD have also improved the across wafer and wafer-to-wafer thickness uniformity of SiNx through modifications to the showerhead, a part of the upper electrode assembly that ideally distributes an even gas flow over the entire diameter of the process chamber.  Initially, a lack of high uniformity in the SiN films on the wafers appeared to follow no pattern, but rather, was quite random. Further investigations at Plasma-Therm identified an issue with the showerhead. This led to a change in the manufacturing process for this part, plus the establishment of a final quality control procedure that drew on the findings of the process testing of the showerheads.

This effort has driven down the level of non-uniformity that is to be expected in SiN films deposited by the Plasma Therm tools beyond what was originally specified by RFMD. Typical thickness non-uniformities on the order of ± 3-4 percent have been the norm but the lower the uniformity the better and the ‘magic’ ± 1.5 percent target is the goal for the future.

The improvements made by Plasma-Therm have driven down the non-uniformities, and today, it is not uncommon for the non-uniformity of the SiN process to be less than ±2 percent; and the uniformity is often significantly better than this. The showerhead, along with the chamber walls and internal chamber fixturing, is a major source of particles in any PECVD deposition system. To minimise these particulates, it is important to heat as many surfaces within the chamber. This leads to better adhesion of deposited material and less flaking or de-bonding of the deposited SiNx from the chamber fixturing. Parts that are heated include the upper electrode. After every in-situ clean, the showerhead is returned to its original manufactured operating condition.

Initially, the benefits of in-situ cleaning went unnoticed, due to the relatively high frequency of placement faults, which had to be addressed by opening the chamber. According to the initial preventative maintenance schedule, an open chamber clean was to take place every six months, but it made sense to also perform these cleans every time the chamber had been opened to address an indexer fault. The frequency of these faults varied from tool to tool, and in some cases took place every two-to-three months.

It is hard to know how long an in-situ chamber clean should be done, and this inevitably leads to ‘overkill’ in the cleaning process. This is very inefficient, because it reduces tool availability. One way to address this, which has been pursued through the Plasma Therm-RFMD partnership, is to turn to Optical Emission Spectroscopy (OES). This provides clear, precise detection of when all the material has been removed (see Figures 3 and 4).

Figure 3. Optical emission spectroscopy can determine when the etching of SiN from the chamber is complete



Figure 4. The EndPointWorks optical emission spectroscopy trace showing endpoint for a chamber clean. The x-axis is time and the Y (left) is signal amplitude.  The Y (right) axis is a differential that is a calculated part of our endpoint software.  This trace includes a cleaning step, plus chamber conditioning to prepare the chamber for further wafer depositions

A common approach to removing SiN is to etch with a fluorine-based gas. This type of approach is used at RFMD, within-situ cleaning involving a mixture of SF6 and N2O – the latter acts as a source of oxygen that bonds with the free sulphur to increase the etch rate. Cleaning is typically carried out after 1.5 µm of SiN deposition, using a process that takes about an hour to both clean (end pointed by OES) and carry out a subsequent post clean ‘conditioning’ deposition.

Thanks to improvements to the in-situ cleaning process, the intervals between open chamber cleans have been increased. In 2010, RFMD’s engineers decided that it was acceptable to carry out the open chamber cleans just once every year, and in late 2011, this interval was extended to two years. These intervals have increased due to the improvements and cooperation by Plasma-Therm and RFMD to resolve the issues that required a continuous, random opening of the chambers.  Through 2012 and the beginning of this year, the LAPECVD tools are continuing to perform at a very high level, and by the first quarter, 2013, uptime over the entire toolset had risen to 98.6 percent. This success is the fruit of a relationship that has been built on trust and has now been going for nearly 20 years. When trust is established, candid and fruitful discussions ensue and equipment performance is improved, benefitting all concerned.  Simply put, working together works wonders!