IEEE Robotics & Automation Magazine - June 2023 - 32

10
15
20
25
30
35
40
45
50
5
00:00:00
00:12:33 00:25:07 00:37:41 00:50:41 01:02:48
Elapsed Time (Granularity: 1 s)
01:15:22 01:27:55 01:40:29 01:53:03 02:05:37
192.169.12.11 CPU 192.169.12.13 CPU 192.169.12.14 CPU 192.169.12.17 CPU
192.169.12.18 CPU
192.169.12.20 CPU 192.169.12.21 CPU 192.169.12.22 CPU 192.169.12.81 CPU (×10) 192.169.12.84 CPU (×10)
FIGURE 6. CPU consumption under scenario C.
TABLE 5. Test Index for FDC-on-Hadoop System.
INDICATORS
Request success ratio
Peak memory
SCENARIO A SCENARIO B SCENARIO C AVERAGE
100.00% 100.00% 100.00% 100.00%
6.50%
Average response time (ms) 375
Total test mime (hh:mm:ss)
5.70%
538
0:47:08
1:36:29
4.60%
515
1:59:11
For scenario C, the CPU utilization is under 40% for most of
the time and its peak value is 44.6%. Notice that the peak value
of CPU utilization is an important indicator showing whether
HE-FDCS can run under a controllable and safe condition. From
the test results of the CPU consumptions, the CPU utilizations
under different scenarios do not reach the upper bound (i.e.,
100%). This implies that HE-FDCS can run safely and stably.
The other test results under different scenarios are summarized
in Table 5. In Table 5, the request success ratio indicates
the ratio of successful interface requests response to the total
requests. For scenarios A, B, and C, the request success ratio
has reached 100%. Also, the peak values of memory utilizations
under scenarios A, B, and C are 6.5%, 5.7%, and 4.6%,
respectively. The test purpose of memory utilization is similar
to CPU and both of them are to test if HE-FDCS performs
safely and stably. For scenarios A, B, and C, the memory utilization
is at a low rate. Furthermore, the average response time
is to test how long HE-FDCS takes from the data-processing
layer and FDC core back-end module at the FDC layer. The
test result is 489 ms under the three scenarios, which is a very
imperceptible time and acceptable to HD-FDCS. Therefore, it
follows from the test results shown in Figures 4-6 and Table 5
32 IEEE ROBOTICS & AUTOMATION MAGAZINE JUNE 2023
5.00%
489
--
that HE-FDCS can perform well under
the manufacturing conditions.
Besides, the current FDC system is
isolated from other software systems,
such as MES, MCS, and APS. The
development of the Hadoop ecosystem
in semiconductor manufacturing
enables data sharing among HE-FDCS
and other CIM systems in the future.
This breaks the boundaries of CIM systems
to enhance the production efficiency and operational cost
reduction with the application of AI-based techniques, IIoT,
cloud computing, etc.
CONCLUSION
As a kind of the most advanced and complex manufacturing
systems, semiconductor manufacturing systems have deployed
software and IoT technologies such that large amounts of data
are collected. The FDC system is a core software system in
CIM in semiconductor fabs. Various data should be input into
the FDC system to observe the healthiness of manufacturing
tools. However, due to the data-processing capacity limitation
of the current FDC system, the system fails to deal with large
data sets such that the diagnosis efficiency is limited, and
undesired events cannot be detected in time. This work conducts
a study on a new FDC framework based on a Hadoop
ecosystem (i.e., HE-FDCS) to cope with the data-processing
limitation issue and improve the diagnosis efficiency. For HEFDCS,
a Hadoop ecosystem is used for data storage, transfer,
processing, and management, while the business logics are left
in the FDC system. In this way, the data stored in the Hadoop
ecosystem can also be used for other software systems if such
Performance Metrics

IEEE Robotics & Automation Magazine - June 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2023