This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
TECH SemiconductorÂ is aÂ subsidiary company of Micron, producer and seller of Dynamic Random Access Memory (DRAM) and NAND logic gates. Micron is one of the world's leading semiconductor companies. Its DRAM, NAND, and NOR Flash memory are used in everything from computing, networking, and server applications, to mobile, embedded, consumer, automotive, and industrial designs. Micron's products are built on decades of design and manufacturing expertise and we continually build on past experience to strengthen the quality of our products and the reliability of our manufacturing processes.
One of the critical processes in semiconductor DRAM product is epitaxial silicon deposition (a chemical vapor deposition process). It uses DCS (DiChlorosilane) gas as the silicon source while H2 gas is used as a carrier gas. In order to maintain good epi deposition uniformity across wafer, epi chambers are kept at low pressure of 40Torr. A vacuum pump is used to maintain the required chamber pressure.
However, epi chamber pump requirement is slightly different from pumps used in other processes. Due to high flow rate of DCS at 40slm and H2 being a light carrier gas compared to other processes, TECH often encounters early pump failures due to high work load on the pump.
Micron Technologies has been growing in DRAM and NAND market share recently thereby making it susceptible to frequent failure in manufacturing output. Pumps are a key part of epi deposition process, as they are used to maintain the chamber pressure at 40Torr for process uniformity. At present, frequent failures were observed in the epi process pumps which are not acceptable with a low MTTF.
Therefore we desire to study the existing operation of the pump and identify its failure modes. The predominant failure mode will be found through a well-known comparison tool called "pumping efficiency curve" comparison method.
Once the predominant failure modes are found, we could make changes to the pump parameters and design (if necessary) to improve its performance. Once the design is improved, we would subject the unit to an ALT to measure its reliability after the change. Reliability is measured in terms of Mean Time to Failure (MTTF).
Introduction to Accelerated Life Testing (ALT)
ALT is the process of determining the reliability of a product in a relatively short period of time (usually weeks or months) by accelerating the operating environment conditions. ALT is also good for finding dominant failure mechanisms and is usually performed on individual assemblies rather than full systems. ALT is also frequently used when there is a wear out mechanism involved.
The practical steps of accelerated life testing experiment are the following:
(1) Allocate a number of components to be tested at each stress level.
(2) Carry out the testing.
(3) Estimate the distribution function or any quantity of interest for each stress level.
(4) Predict the distribution function or the quantity of interest for the normal stress level through "extrapolation".
In order to set up an ALT, we must know several parameters, including but not limited to:
Length of Test
Number of Samples
Goal of Test
Acceleration Factor Calculation
Slope of Weibull Distribution or Beta parameter (Beta<1 indicates infant mortality, Beta>1 indicates wear-out)
The ebara pump under consideration is a vacuum pump used to maintain the epi chamber pressure by regulating the exhaust of DCS with the help of H2 as a carrier gas.
The normal operating conditions of the pump are as follows:
Peak Speed - 950 m3h-1
Ultimate Vacuum - 1.5 x 10-3 Torr
Typical shaft seal nitrogen flow - 4 slm
Inlet Connection - ISO100
Outlet Connection - NW40
Typical cooling water flow at 15 psi pressure drop - 240 l h-1
Weight - 430 kg
Power input at ultimate - 3.5 kW
Rated motor power - 5.1 kW
The Picture below shows a typical Ebara pump used for epi chambers.
Objective of Study
To study the existing operation of the pump and identify its failure modes.
Identify the predominant failure mode through a well-known comparison tool called "pumping efficiency curve" comparison method.
Make changes to the existing system if necessary to increase the pumping efficiency.
Use ALT to reduce the time to failure by testing at higher stress levels and thereby determining the MTTF under desired conditions after the changes.
The following is the cross section diagram of existing EBARA vacuum pump unit:-
The following is the fault tree for the Ebara vacuum pump:-
Common Failure Modes
The following table shows the failure mode and effects for the Ebara pump:-
Part wear out
Particles/ By-product back streaming
Controller failure/ Facilities incoming
Chocking in the line
Pumping efficiency reduction
N2 gas flow
Gauge failure/ Facilities incoming
Pumping efficiency reduction
ELB Circuit/ Control circuit
Arcing / Short Circuit
Weak Point Analysis
Based on the above failure possibilities, it is important to identify the predominant failure mode of the vacuum pump unit for EPI process. In order to identify the predominant failure mode, we use a well-known comparison tool called "pumping efficiency curve" comparison method.
As per expert's opinion, failures occurring due to electrical components are very minimal compared to the occurrence of mechanical failures. Pumping efficiency curve check was performed on 4 mechanical parameters as mentioned below:
In this Pumping efficiency comparison method, we plot H2 flow (sccm) on X axis and pressure (Torr) on Y axis, with the following measuring parameters at various levels to determine the most sensitive parameter.
N2 flow (simulate the dilution effect on light gases like H2 to relieve stress on pump)
Pumping speed (to simulate for fore line/ pump line chocking)
Pump cooling water (to simulate the effect of surrounding temperature on pump)
Availability of check valve (to simulate if efficiency is affected by back streaming)
The current system operates at the following baseline values,
N2 flow at 4 psi
Pumping frequency at 60 Hz
Cooling water temperature (30Â°C)
Pumping efficiency curve by N2 flow:
The following table shows the H2 flow vs. pressure on the pump at various N2 pressures.
N2 pressure \ H2 Flow(slm)
N2 flow (4psi) Baseline
N2 flow (6psi)
N2 flow (8psi)
N2 flow (10psi)
N2 flow (11psi)
N2 flow (12psi)
Inference from the curve:
Pumping efficiency curve for N2 flow is substantially sensitive for values above baseline.
N2 flow is found to be optimal at above 10psi.
Pumping efficiency curve by Pumping speed simulating line chocking :
The following table shows the H2 flow vs. pressure on the pump at various Pump Frequencies.
Pump frequency (50Hz);
Pump frequency (60Hz); Baseline
Pump frequency (80Hz)
Pump frequency (100Hz)
Inference from the curve:
Pumping efficiency curve for various pumping speed is substantially sensitive for values above baseline.
Pumping frequency is found to be optimal at above 100Hz.
Pumping efficiency curve by Cooling water temp & check valve:
The following table shows the H2 flow vs. pressure on the pump at various cooling water temperature.
Cooling water (30degC); Baseline
Cooling water (32.5degC);
Cooling water (35degC)
Baseline without check valve
Inference from the curve:
Pumping efficiency curve for both cooling water temperature & check valve is found to be less sensitive for valves above baseline.
The predominant failure modes in decreasing order were identified to be
N2 pressure for dilution (1st priority)
Pumping speed (Fore line/ pump line chocking (2nd priority))
The following Design Changes were made
Increase the N2 pressure from baseline of 4psi to 10psi for better pumping efficiency.
Change the pump frequency to 100Hz (cost incurred with minimal change to pumping efficiency).
Now, we run the ALT test to measure the reliability of the system. ALT is used to accelerate the failure of the system. Failure is accelerated by applying high stress levels, but at the same time the high stress should not excite new failure modes that would have not occurred at normal operating conditions.
Due to the nature of the operating environment to which the pump under consideration will be subjected to, we found that the pump in use fails due to the inability of the pump to pull the light DCS H2 N2 mixture at baseline conditions.
We selected three stress levels:
Lower / Middle/ Upper Stress: 1 psi / 2 & 3 psi / 4 psi
Total number of sample available: 15
Since higher stress has higher probability of failure, we unequally allocated the samples as
Lower / Middle/ Upper Stress: 3: 8: 4
N2 pressure (psi)
Life time (Hrs)
Stress - Life graphs were plotted and the expected life time at the normal operating stress was extrapolated.
Plotting the results we get the following equation matching the relation between stress and life time,
Life time = 13915 e-1.3518*Stress
Life time measured in hrs, Stress measured in psi.
Using this equation, we could find the MTTF at normal operating conditions of stress level at 0.1 (i.e. N2 pressure at 10psi).
MTTF (10 psi) = 13915 e-1.3518*Stress = 12155.6hrs = 16.88 months (Approx)
At manufacturer's optimal pump condition, stress = 0.1, MTTF = 16.88 months.
At manufacturer's specification limit, stress = 0.5, MTTF = 9.83 months.
Under normal working stress levels, the pump after the proposed changes has a mean life time of 12155.6 hrs. The below table shows the MTTF.
As shown in the table above, MTTF was found at standard operating pressure of 30Torr for epitaxial silicon deposition chambers. MTTF was found based on failure data obtained from a group of 15 samples. 12 epi chambers Ebara pump samples were analyzed for its average failure occurrence over the past 4 year's time frame and tabulated below.
Pump failures snumber
MTTF (in hrs)
Reliability for the pump was calculated based on the failure data collected as above.
Exponential distribution curve
In usual industrial practice, exponential distribution is commonly used to fit the life time distribution and to calculate reliability. We tried using exponential distribution to fit the data, but lifetime curve doesn't fit the failure data exactly.
The following graph shows the exponential distribution used to fit the failure data. As seen, the failure data points are far away from the exponential line.
Coefficient of Determination = 0.6627 (Low - not acceptable).
Weibull distribution curve
The Weibull distribution better fits failure data as seen above.
Coefficient of Correlation = 0.9676 (Good Fit)
Scale Parameter Î± = 1.3459 * 104
Shape Parameter Î² = 5.3662
Shape Parameter Î² >> 1 indicates increasing failure rate.
Weibull distribution which better fits the data has a shape parameter beta >>1. This proves that the failure rate is increasing; hence exponential distribution would not be a perfect model, as it can't be used for serious deviations from constant failure rate.
Probability Density Function
Failure Rate vs. Time
Since Î² >>1, failure rate increases with time, which is evident from the following plot.
The Reliability vs Time plot was used to predict the reliability of Ebara pumps for the required life time of the pump.
The pump under consideration is expected to work for a minimum of 17 months. The design improvements were made to make sure that the pump works until the desired life time.
Hence we calculated the reliability of the improved pump at 17 months assuming Weibull distribution for the lifetime distribution.
Reliability (T = 17 months) = 0.6 (Approx), which means that 60% of the pump would still be working after 17 months.
HALT test was used to improve the design of the transformer and ALT test was conducted to measure its reliability. The design improvements could be implementation in all future transformers to be produced. In order to test each unit being manufactured, Highly Accelerated Stress Screen (HASS) could be executed at the end of the production, to make sure that the product would work properly within the specification limits. This would reduce warranty and repair costs by reducing external failures after the product has been sold to the customer. Any fault in construction would reflect in HASS through infant failure of the product during the HASS test.