An improved pattern match method with flexible mask for automatic inspection in the LCD manufacturing process

Abstract

This research examines the electrical conductive particles of anisotropic conductive film (ACF) during the laminate manufacturing process for liquid crystal displays (LCD) and adopts an improved pattern match method to apply on-line automatic and relevant measurement inspections.

The focus of this research is describing and identifying feature images. We aim to improve the traditional pattern match methods including the gray scale designs, adaptive pattern matrix, and adaptive feature weight pattern to reduce the system function errors and to enable more efficient and quicker pattern searches and matches for full-size images.

Introduction

ACF is a key material that allows LCD panels and its peripheral circuits to achieve vertical conductivity. This research adopts the hardware for machine vision and pattern identification and match technology to conduct multiple match functions on an ACF laminate test image of an LCD panel to identify the distraction and quantity of conductive particles inside each bump and to inspect and measure the movement of conductive particles after lamination.

Its purpose is to inspect and measure the movement of ACF conductive particles and ensure the correct position of the ACF particle layer with chip mark and ITO glass substrate with cell mark and to calculate whether the movement is within the allowable range. As we can see in Fig. 1, the measurement principle looks for the central point of a cell mark and bump center of conductive particle distribution to calculate the horizontal ΔX movement and vertical ΔY shifting.

Fig. 2 shows the definition of the pattern match system. Discrepancy of test image and objective pattern is D_ad. The pattern match algorithm first defines one object image and then calculates the variants of the image information groups R, G, and B to search an identical block image. We assume there is a subdivided block that includes three pieces of surface description. R_i, G_i, and B_i represent the mean of the color space expression. In addition, D_i = [R_i, G_i, B_i]^T shows the individual description of each divided block where D is from the whole divided block; that is, D = [D₁, D₂, … , D_n], where n is the number of divisions in the whole block. We assume one standard pattern block as D′ and $D^{\prime }=[D_1^{\prime }\text{,}{D}_2^{\prime }\text{,}\dots \text{,}{D}_n^{\prime }]$, $D_i^{\prime }=[R_i^{\prime }\text{,}{G}_i^{\prime }\text{,}{B}_i^{\prime }{]}^{\mathrm{T}}$.

The discrepancy relationship of each divided block image is shown in Eq. (1) to compare the differences between the standard block pattern and the moved block pattern:

$$D_d=|D-D^{\prime }|=\left[\begin{array}{cccc}\hfill |R_1-R_1^{\prime }|\hfill & \hfill |R_2-R_2^{\prime }|\hfill & \hfill \dots \hfill & \hfill |R_n-R_n^{\prime }|\hfill \\ \hfill |G_1-G_1^{\prime }|\hfill & \hfill |G_2-G_2^{\prime }|\hfill & \hfill \dots \hfill & \hfill |G_n-G_n^{\prime }|\hfill \\ \hfill |B_1-B_1^{\prime }|\hfill & \hfill |B_2-B_2^{\prime }|\hfill & \hfill \dots \hfill & \hfill |B_n-B_n^{\prime }|\hfill \end{array}\right]=\left[\begin{array}{cccc}\hfill \mathrm{\Delta }{R}_1\hfill & \hfill \mathrm{\Delta }{R}_2\hfill & \hfill \dots \hfill & \hfill \mathrm{\Delta }{R}_n\hfill \\ \hfill \mathrm{\Delta }{G}_1\hfill & \hfill \mathrm{\Delta }{G}_2\hfill & \hfill \dots \hfill & \hfill \mathrm{\Delta }{G}_n\hfill \\ \hfill \mathrm{\Delta }{B}_1\hfill & \hfill \mathrm{\Delta }{B}_2\hfill & \hfill \dots \hfill & \hfill \mathrm{\Delta }{B}_n\hfill \end{array}\right]$$

From Eq. (1), we get Eq. (2)

$$D_{\mathit{ad}}=\sum _{i=1}^{i=n}(\mathrm{\Delta }{R}_i+\mathrm{\Delta }{G}_i+\mathrm{\Delta }{B}_i)$$

D_ad refers to the total number of differences that exist in the image between surface spaces R, G, and B of the whole block and the standard pattern. It is also used to evaluate the quantified value of the discrepancy, one of the descriptive features for evaluating proximity in this block.

The definition of a flexible mask system and relevant matrix parameters can be expressed as follows:

Test image I_i(x, y) at the size of M × N, 1 ⩽ i ⩽ 3; (I₁(x, y), I₂(x, y), I₃(x, y)), represents the gray scales (R, G, B) of the coordinate point (x, y).

Objective pattern T_i,j,k(p, q), at the size of P × Q, 1 ⩽ i ⩽ 3, 1 ⩽ p ⩽ P, 1 ⩽ q ⩽ Q; T_i,j,k(p, q), refers to the pattern coordinate system of (j, k) based on the comparative first point of the pattern; (T_1,j,k(p, q), T_2,j,k(p, q), T_3,j,k(p, q)) shows the gray scale of the coordinate point of the patterns R, G, and B, respectively.

The pixel of objective images has a gray scale (R, G, B) of (N₁, N₂, N₃) and n refers to number.

Feature weight matrix is shown as W(p, q) at the size of P × Q,

$$W(p\text{,}q)=\left[\begin{array}{ccc}\hfill W(1\text{,}1)\hfill & \hfill \cdots \hfill & \hfill W(1\text{,}Q)\hfill \\ \hfill ⋮\hfill & \hfill \ddots \hfill & \hfill ⋮\hfill \\ \hfill W(P\text{,}1)\hfill & \hfill \cdots \hfill & \hfill W(P\text{,}Q)\hfill \end{array}\right]$$