1063-7397/02/3102- $27.00 © 2002 MAIK “Nauka /Interperiodica”
Russian Microelectronics, Vol. 31, No. 2, 2002, pp. 130–136. Translated from Mikroelektronika, Vol. 31, No. 2, 2002, pp. 152–160.
Original Russian Text Copyright © 2002by Doroshevich, Popov, Strizhkov.
It is typical of the modern semiconductor industry to
manufacture ICs on order. In some cases, production is
organized on a small-batch basis and has intermittent
If intermissions occur, the manufacturer has to carry
out comprehensive inspection and readjustment of the
process in order to maintain the required quality of cur-
rent products or to change to new ones. In Russia, the
problem of process-quality control in small-batch,
intermittent production has become particularly press-
ing for small manufacturers of LSIs and VLSIs.
The above problem could in principle be solved by
statistical process control (SPC). However, the inade-
quate amounts of statistical data available in actual sit-
uations prevent us from using well-known tools, such
as process-capability indices and control charts. We
therefore devised and tested a technique that allows one
to apply statistical methods to small-batch, intermittent
production in the microelectronic industry. The tech-
(1) a method to estimate statistical characteristics
when the sample size changes from batch to batch and
(2) a method to estimate the variability bounds of a
monitored variable in the case of small sample sizes.
This paper describes our approach and illustrates it
with real-world data.
PRINCIPLES OF SPC IN SMALL-BATCH,
SPC distinguishes between ordinary and special
causes of variability in a monitored quantity with the
aim of eliminating the latter. Special causes can be
detected by means of process-capability indices,
Shewhart charts with warning and action limits, etc.
Under the manufacturing conditions concerned, we
have to modify the conventional techniques of setting
limits in a control chart. It is advisable that the enter-
prise standard meet these requirements:
(1) It should deﬁne the conditions that would allow
one to combine data concerning the same monitored
variable for different versions of a product. For exam-
ple, the data on different versions may be combined if
the respective processes obey the same tolerance limits
or run under the same conditions, the wafers under pro-
cessing are positioned in the same manner, etc.
(2) It should allow one to accumulate an adequate
amount of statistical data by increasing, to an accept-
able extent, the number of test points on a wafer and the
number of wafers inspected in a batch.
(3) It should cover the in-process estimation of sta-
tistical characteristics from measured values of critical
If no process history is available, the process status
at a speciﬁc step can be evaluated for each batch. The
evaluation should be carried out when an intermission
occurs and in the next cycle of production. It should
involve checking the estimated variability bounds 
and the individual values of a monitored variable
against a relevant standard. If the result is negative, one
should ascertain the deviation of the process conditions
for the current batch from those for batches that passed
the inspection. On this basis, process status should be
evaluated and a decision should be made about what
measures are necessary.
If the result is positive for ﬁve to ten batches, the fol-
lowing procedures should be executed:
(1) Based on the data collected over all of the
batches, compute the arithmetic mean and the rms devi-
(2) Using the statistical characteristics, evaluate the
indices of process spread, accuracy, and stability
according to the formulas given by the GOST RV
20.57.412-97 national standard.
(3) Set tentative control limits in the control charts
for arithmetic mean and rms deviation according to a
relevant branch standard .
The control limits should be set for batch groups
such that every batch in a group would give the same
Statistical Process Control in IC Manufacture: A Technique
for Small-Batch, Intermittent Production
K. K. Doroshevich, V. N. Popov, and S. A. Strizhkov*
Received October 17, 2001
—A technique of statistical process control for small-batch, intermittent production is presented. It
rests on the estimation of upper and lower variability bounds for monitored variables from a limited amount of
PROCESS QUALITY CONTROL