ABC Semiconductor Case Study: Improving CMP Yield
ABC Semiconductor Case Study:
Improving CMP YieldMicrel Gas Cylinder
Case Study
1/14/08October 2008
SUMMARY –I think this is too long I think for a summary and could be divided into sections for easy skimming – it should be past/current tense as I’ve revised
§ Micrel is a publicly traded semiconductor company with its fab operations based in San Jose, CA.
CHALLENGE
Most of the fabMicrel’s fab was constructed in the 1980’s and the facilities and related automation technology is representative of that periodand is considered a legacy fab without current automation technology. They constantly assess technology upgrades to reduce cost and improve productivity to remain competitive. Managing the use of process gases, which is labor intensive and prone to gas shortages and/or waste, was identified as a key area of potential improvement.
§
§ Micrel is constantly assessing technology upgrades to reduce cost and improve productivity to remain competitive with current day practices.
§ Managing the use of process gases, traditionally labor intensive and prone to gas shortages and/or waste, has been identified as a key area of potential improvement.
§ Management considered installing transducers and monitoring equipment to automate gas rounds, but found the cost of installation, wiring, and disruption to ongoing operations prohibitive with poor payback.
SOLUTION
In November 2007, Micrel evaluated and decided to proceed with the installationinstalled of new automation technology which did not incur any disruption to ongoing processes required minimal installation cost. The solution, a non-invasive wireless sensor which “clips-on” to existing gauges and/or transducers, is one-third the cost of alternative solution of traditional transducers installation.
RESULTS
§ The new technology is a non-invasive wireless sensor which easily “clips-on” to existing gauges and/or transducers, with a resulting installed cost only one third of traditional transducers.
§ Based on initial resultsreturn on investment - baesd on how much time?, Micrel estimates annual savings to be in the range of $250,000K, which correlates to an investment payback time of approximately 6six months.
The Challenge:
ABC Semiconductor operates an 8-inch fab in Texas with an average of 20,000 wafer starts per month. As part of the production process, it uses twenty seven Chemical Mechanical Planarization (CMP) tools manufactured by Westech, model 472 (circa 1995 vintage).
The CMP tools are used at various stages of the fabrication process, and their yield has been averaging well under 90%. At a cost of approximately $1,250 per scrapped wafer, the impact to profitability is huge.
ABC Semiconductor convened a Kaizan team to examine the CMP process to try to reduce the scrap rate. The team identified five key unmonitored variables which they believe may be affecting the yield rate.
§ Polishing head downward pressure
§ Platten up pressure
§ Platten rotation torque
§ Slurry flow rate
As a company which takes prides in its Six-Sigma culture, ABC Semiconductor wanted to measure and capture these process variables on their Fabguard fault detection system so they may analyze it using statistical process control tools and establish control limits. They believe that once they identify the variations which occur, they can take proactive steps to address the root causes and improve consistency.
Unfortunately, unlike modern day tools, the Westech units do not provide the proper data outputs to send to the Fabguard system (the SECS/GEM interface is very limited, typical of machines of that vintage). The tools do have gauges and transducers installed, but they can only be manually read by operators. There appeared to be no easy way for ABC Semiconductor to capture the process data they need short of buying new tools.
Cypress Envirosystems’s Solution:
Cypress Envirosystems proposed using its Wireless Gauge Readers to read existing pressure gauges for downward head pressure and platten upward pressure. Cypress also proposed using its Wireless Transducer Readers to capture the current draw of the motor to infer the platten rotation torque. Finally, Cypress proposed using a clamp-on ultrasonic flow meter together with a Wireless Transducer Reader to measure the slurry flow rate.
ABC Semiconductor decided to install the Cypress solution on two of its CMP tools as a trial. The installation was completely non-invasive (i.e. did not involve disrupting or modifying any flow lines or electronic element of the tool). The non invasive nature avoided any need to requalify the tool, and the installation took less than three hours.
Two identical CMP tools were selected for the initial trial so that both the variability within a single tool, and variability between different tools, can be determined. Data was collected at a sampling interval of 20 seconds and fed to the Fabguard system using a standard OPC communications protocol over their factory LAN.
Figure 1 – Circa 1995 Westech 472 CMP Tool
Figure 2 – Non-Invasive Mounting for Wireless Gauge Reader
The Results
Within less than two weeks, the Fabguard system was able to collect sufficient data to determine that variability in the polishing head downward pressure and the slurry flow rate were the main contributors to process variation and ultimately lower yield.
ABC Semiconductor developed a corrective action plan which involved procedural changes including monitoring slurry filter change-outs and better regulation of the compressed air sub-system. The yield rate improvement was conservatively estimated to be about 1%. Further improvements are anticipated as more data is gathered and analyzed.
As a result of the trial, ABC Semiconductor plans to deploy the Cypress monitoring system to all twenty seven CMP tools at a cost of approximately $235,000. Related corrective action measures would require an additional $250,000, for a project total cost of $485,000.
The project would achieve a yield improvement of 1%, which translates into annual savings of $3,000,000. The project payback period is less than two months.
Situation/Background
As new semiconductor fabs continue to be built globally, there are hundreds of ”mature” fabs in the United States and thousands worldwide that are still the workhorses of the industry. These legacy fabs, which are often twenty years or older, produce millions of chips for a huge array of electronics every day. And while many newer fabs have automation features built in to enable peak efficiency and performance, older fabs seek new solutions to increase productivity, reduce downtime, and decrease operating costs.
One such company is Micrel, based in San Jose, California, a leading IC solutions manufacturer for the analog, Ethernet and high bandwidth markets. The company manufactures high performance analog, power, advanced mixed-signal and radio frequency semiconductors, high speed communication, clock management, and Ethernet switch and physical layer transceiver integrated circuits. End markets served include cell phones, portable and enterprise computing, enterprise and home networking, wide area and metropolitan area networks and industrial equipment. Founded in 1978, Micrel has been profitable for 25 of its 26 years with revenues of $280M and regional sales and support offices, sales offices, distributors and technology design centers throughout the Americas, Europe and Asia.
Challenge: Gas Level Monitoring
For Micrel, it is critical to minimize unplanned downtime at its core San Jose fab. At this facility, which houses 500 semiconductor manufacturing tools, wafer production requires hundreds of process recipes that use various combinations of gases to produce desired chips. It is critical that the required gases are provided for each wafer process step or the output will be unusable and system downtime occurs as a result. Essentially, semiconductor production hinges on the availability of these gases so no chances can be taken -- ever. The gases themselves, supplied by about 300 different cylinders (see Fig 1), are a precious commodity, ranging in cost from $800 to $17,000 per cylinder. – I am not sure it is necessary to show what the cylinders look like
Fig. 1 – Micrel Gas Cylinders
For years, like many older fabs, monitoring gas cylinder levels has been a manual process. At Micrel, once every 12-hour shift, an employee armed with paper and a clipboard went to each gas cylinder location to manually check the pressure regulator gauges and/or the weigh scale depending on the gas type. This process took about 4 hours each time or about 8 manlabor-hours per day. By 2001, a bar code system was employed to identify each gas cylinder, but an employee was still required to manually read and record each data point, so the process was still error- prone (we haven’t discussed errors…should do that before we mention it here?) and labor-intensive.
In addition to monitoring gas levels manually, to further reduce risk of low or empty cylinders, gases were typically? It is or isn’t I would think… replaced on a regular schedule – even if a cylinder may be a quarter or even a third full, just to be safe. ß-this is redundant. Despite these precautions, gas supply disruptions still periodically occurred due to unforeseen circumstances such as cold weather impacts which can condense a gas. Any other circumstances? Such disruptions often resulted in lost yield and process downtime, which translated into additional operational expense.
Altogether, these labor, gas, and downtime related expenditures, year after year, were continuing to add up and needed to be addressed to keep the fab competitive. Operations management sought new ways to better utilize manpower, ensure gases were expended before replacing bottles, and proactively detect situations which cause downtime.
Evaluation of Traditional Automation Solution Too Costlyshow them don’t tell them
One solution that was explored was to replacereplacing the manual gauges with all-new transducer based gas panels, which would automate the monitoring of gas levels. However, making this change would require Micrel to stop the affected processmanufacturing production? for up to a few days – I thought it was a lot longer? Guess not. while it reconfigures gas piping and retests for leaks and safety. For some cases, thThis can also e replacement can in turn triggerinvolve ana even broadermuch larger rework of the entire wiring system in order to conform to the latest safety and fire codes – required by government regulations when a legacy subsystem is significantly modified.
Micrel chose not to implement this solution due to the prohibitive cost of the equipment, the installation and rework labor, and most of all, the associated downtime and lost production, which together added upaltogeter was to over $3,500 per gas cylinder.
The Cypress Systems Solution
In 2007, Micrel learned of new “wireless gauge reader” and “wireless transducer reader” why the quotes? technologies that can provide automated gas cylinder monitoring without the high cost and disruption of installing traditional transducer based panels. The wireless gauge readers non invasively do not require the existing manual regulator gauges to be removed. Instead, they simply “clip” onto the front of the legacy manual gauge, and transmits the reading wirelessly to a central server (Fig. 2). Similarly, the wireless transducer readers simply attach to existing gas weigh scales without the need to remove or replace the scale. Pre-set aAlarms to notify of low or critically low gas levels? may be programmed to alert operator stations, pagers, or cell phones when required.
Figure 2 – Wireless Gauge Reader Non-Invasive Mounting
Micrel electeddecided to implement thisthe Cympress solution because it incurreds no process downtime, requireds minimal installation labor and training, and does not involveno retesting and revalidation for leaks. The resulting installed cost at $1,250 per point (list price) is only aboutjust 35% of the cost of traditional transducer panels (Figure. 3).
Figure. 3 – Comparison of Installed Cost per Point
for Traditional Transducers vs. Wireless Gauge Reader
Results
In Q4tehe fall of 2007, Micrel decided to install a mix of 100 wireless gauge readers and wireless transducer readers from Cypress Systems to monitor its most critical process gases, along with the Cypress Systems’ Blue Box Receiver (BBR-100), which sends the data to Micrel’s existing network and operator stations. Each data point required between 10 and 30 minutes to install, and did not involve breaking any pressure seals anywhere in the gas system or any process downtime. of any sort.
Maybe we should have bullet points for results since there are a number of them? It would be easier to read.
As a result
· The Cypress Solution enabled of this system, Micrel was able to reduce the manpower devoted to making gas rounds by one full-time equivalent technician, who could then may then be redeployed to other more pressing tasks.
· In addition, Micrel estimates thata gas consumption wouldto be reducedreduction by about 10% in the first year. Finally,
· even during its first month of operation, the new system was able to detect trends and fluctuations in gas pressure due to unseasonably cold weather in January and helped to avoid future downtime and yield loss. ßthis doesn’t seem to go with the other points and is a bit awkward.
· Altogether, Tthe combined benefitsreturn on investment are estimated to be in the range of $200,000-250,000K per year (Figure. 4). ComparedWhile with the cost of installing the system at $125,000K, the payback period is only 6 to-8 months (Figure. 5).
Figure 34. – Savings Derived from Wireless Gauge Reader and Wireless Transducer Reader Solution – I can’t revise this figure but a few comments - do we really need to have all this explanation? Such as “basis” and “comments”? You’ve already explained it in the text and I think you might get into too many details, they may have a different situation and you are locking yourself into this one. The opint is to give an example of savings. I would eliminate at the least the “Comments” call “basis” “Cost Basis”. We also need a title for Labor Savings, etc. How about “Cost Area”?
Figure 45. – Payback Analysis for Wireless Gauge Reader and Wireless Transducer Reader