Professional Documents
Culture Documents
Avm in Mesasmc 2010
Avm in Mesasmc 2010
net/publication/224168456
CITATIONS READS
4 285
5 authors, including:
Fan-Tien Cheng
National Cheng Kung University
177 PUBLICATIONS 3,030 CITATIONS
SEE PROFILE
All content following this page was uploaded by Fan-Tien Cheng on 08 September 2015.
– The VM solution must be adaptive Æ Resolved by The VM server, which contains data preprocessing
the dual-phase VM scheme [15]. [17], conjecture model with advanced dual-phase VM
algorithm [15], [17], reliance index (RI) module [16], and
– The VM solution must incorporate VM data quality global similarity index (GSI) module [16], is shown in Fig. 1.
prediction Æ Resolved by the reliance index (RI) Besides the fundamental functions of VM on-line
[16]. conjecturing and VM quality evaluation, the VM server also
possesses the functions of automatic data quality evaluation
– The VM methods need to be improved for and automatic model refreshing [17]. The inputs of a VM
determining the fault detection (FD) factors that server include process data of the current tool and
are the best VM predictors Æ Resolved by the pre-metrology (Pre-Y) data of the previous tool as well as the
global similarity index (GSI) and individual metrology data of the current tool. The Pre-Y data is also
similarity index (ISI) [16]. called Type-2 data in [19] and [20] and may be either actual
– The VM solution must be developed to be or virtual metrology data. The outputs are Phase-I VM (VMI),
re-usable Æ Resolved by the automatic model Phase-II VM (VMII), RI, and GSI values. The kernel of the
refreshing [17]. AVM system is the advanced dual-phase VM algorithm that
will be briefly introduced next.
The RI/GSI/ISI and dual-phase VM schemes have
been successfully verified in Taiwan Semiconductor As mentioned in [15], Phase I emphasizes
Manufacturing Company, Ltd. (TSMC) [16], [18]. All the promptness and Phase II improves accuracy. Observing the
resolutions [15], [16], [17] mentioned above are consolidated right-hand portion of Fig. 2, the Phase-I algorithm starts to
and become the Automatic Virtual Metrology (AVM) system collect the process data of each processing workpiece after the
that has been successfully deployed in Chi Mei conjecture model is built.
Optoelectronics Ltd. (CMO), Taiwan [17], [18]. Therefore, The DQIX algorithm [17] will be applied to evaluate
the AVM system is mature enough and ready to be integrated the quality of the collected process data once process data
into the manufacturing execution system (MES). collection of the said workpiece is completed. If an
This paper proposes a novel manufacturing system abnormality is detected, a warning signal will be sent to the
VMI for FB
AVM VMII for FF
R2R Control R2R Control
Equipment Manager
25 pcs )
)
1 pc )
)