Cross-die interconnection monitor method and cross-die interconnection monitor system capable of extracting cross-die interconnection data for multi-die packages

TWI945613BActive Publication Date: 2026-11-11MEDIATEK INC
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
TW113145695
Authority / Receiving Office
TW · TW
Patent Type
Patents
Current Assignee / Owner
Priority Date
2024-11-14
Filing Date
2024-11-27
Publication Date
2026-11-11
Estimated Expiration
2044-11-26

Smart Images

  • Figure TWG2TB001920358_001
    Figure TWG2TB001920358_001
  • Figure TWG2TB001920358_002
    Figure TWG2TB001920358_002
  • Figure TWG2TB001920358_003
    Figure TWG2TB001920358_003
Patent Text Reader

Abstract

A cross-wafer interconnect monitoring method includes: providing a first wafer and a second wafer; embedding an in-wafer detector into the first wafer to detect a first feature of the first wafer; assigning an inter-wafer detector from the first wafer to the second wafer to detect a second feature between the first and second wafers; and comparing the first feature with the second feature using a neural network to generate a cross-wafer interconnect from the first wafer to the second wafer.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to semiconductor technology, and in particular to a cross-wafer interconnection monitoring method and system. Prior Art

[0002] With the rapid development of technology, various chips and integrated circuits (ICs) are being adopted in our daily lives. Therefore, various electronic applications require high-quality and low-risk ICs. In the silicon test process, in order to provide high-quality and low-risk ICs, abnormal ICs are identified and marked by analyzing the measured test data.

[0003] While outlier ICs can be identified, die-to-die interconnects play a critical role in the performance of multi-die packages. For example, if the impedance of the die-to-die interconnects does not match the target impedance, performance degradation may result. Currently, it is difficult to measure cross-die interconnect data or performance in multi-die packages because fabs only provide wafer acceptance test (WAT) data.

[0004] Therefore, developing a cross-wafer interconnect monitoring system is an important issue. Summary of the invention

[0005] In one embodiment of the present invention, a cross-wafer interconnect monitoring method is disclosed. The cross-wafer interconnect monitoring method includes providing a first wafer and a second wafer, embedding an intra-wafer detector into the first wafer to detect a first feature of the first wafer, allocating an inter-wafer detector from the first wafer to the second wafer to detect a second feature between the first wafer and the second wafer, and comparing the first feature with the second feature through a neural network to generate cross-wafer interconnect data from the first wafer to the second wafer.

[0006] In another embodiment of the present invention, a cross-wafer interconnect monitoring system is disclosed. The cross-wafer interconnect monitoring system includes a first wafer, a second wafer, an intra-wafer detector embedded in the first wafer, a first inter-wafer detector allocated from the first wafer to the second wafer, and a neural network connected to the intra-wafer detector and the first inter-wafer detector. The intra-wafer detector is used to detect a first feature of the first wafer. The first inter-wafer detector is used to detect a second feature between the first wafer and the second wafer. The neural network compares the first feature with the second feature to generate cross-wafer interconnect data from the first wafer to the second wafer.

[0007] These and other objects of the present invention will no doubt become apparent to those having ordinary skill in the art after reading the following detailed description of the preferred embodiment which is depicted in the various figures and diagrams. Simple diagram description

[0008] FIG. 1 is a block diagram of a cross-wafer interconnect monitoring system according to an embodiment of the present invention. FIG. 2 is a block diagram of an inter-wafer interconnect monitoring system according to another embodiment of the present invention. FIG. 3 is an illustration of the neural network inference of cross-chip interconnect data of the cross-chip interconnect monitoring system in FIG. 1 . FIG. 4 is a flow chart of the cross-wafer interconnect monitoring system in FIG. 1 executing the cross-wafer interconnect monitoring method. Implementation

[0009] FIG. 1 is a block diagram of a cross-wafer interconnect monitoring system 100 according to an embodiment of the present invention. The cross-wafer interconnect monitoring system 100 includes a first wafer 10, a second wafer 11, an intra-wafer detector 10a, a first inter-wafer detector 10b, and a neural network 12. The intra-wafer detector 10a is embedded in the first wafer 10. The first inter-wafer detector 10b is allocated from the first wafer 10 to the second wafer 11. The neural network 12 is connected to the intra-wafer detector 10a and the first inter-wafer detector 10b. The intra-wafer detector 10a and the first inter-wafer detector 10b may include a ring oscillator detector. Therefore, the intra-wafer detector 10a and the first inter-wafer detector 10b may be used to detect frequency differences between different wafers. Here, since the intra-wafer detector 10a is embedded in the first wafer 10, the intra-wafer detector 10a may be used to detect a first feature of the first wafer 10. In addition, the first inter-wafer detector 10b may include a first detection circuit 10b1 and a first metal wiring interconnect layer 10b2. The first detection circuit 10b1 is embedded in the first wafer 10. The first metal wiring interconnect layer 10b2 is distributed from the first wafer 10 to the second wafer 11 as a metal loop circuit. Therefore, the first inter-wafer detector 10b can be used to detect the second feature between the first wafer 10 and the second wafer 11. The neural network 12 can be any artificial intelligence (AI) based architecture, such as a convolutional neural network (CNN) or a recurrent neural network (RNN). In addition, the neural network 12 can establish a training model for comparing the first feature with the second feature to infer the cross-wafer interconnection data from the first wafer 10 to the second wafer 11.

[0010] In the cross-wafer interconnect monitoring system 100, as described above, when the intra-wafer detector 10a and the first inter-wafer detector 10b include ring oscillator detectors, the first feature and the second feature may be two frequency signals. For example, the intra-wafer detector 10a may detect a first frequency signal of the first wafer 10. The first inter-wafer detector 10b may detect a second frequency signal affected by the first wafer 10 and the second wafer 11. In addition, the cross-wafer interconnect data from the first wafer 10 to the second wafer 11 may be a resistor-capacitor (RC) interconnect data from the first wafer 10 to the second wafer 11. However, the present invention is not limited thereto. The neural network 12 may infer the RC interconnect data from the first wafer 10 to the second wafer 11 based on the difference between the first frequency signal and the second frequency signal.

[0011] In the cross-wafer interconnect monitoring system 100, the length of the first metal wiring interconnect layer 10b2 of the first inter-wafer detector 10b may be adjustable. For example, the length L1 of the first metal wiring interconnect layer 10b2 may be used to detect the frequency F1 affected by the first wafer 10 and the second wafer 11. The length L2 of the first metal wiring interconnect layer 10b2 may be used to detect the frequency F2 affected by the first wafer 10 and the second wafer 11. Specifically, different lengths of the first metal wiring interconnect layer result in different detection frequencies. Therefore, in the cross-wafer interconnect monitoring system 100, when a plurality of first inter-wafer detectors 10b having first metal wiring interconnect layers of different lengths are introduced, the neural network 12 may detect at least one additional second feature between the first wafer 10 and the second wafer 11. As a result, the neural network 12 may infer accurate cross-wafer interconnect profiles.

[0012] FIG. 2 is a block diagram of a cross-wafer interconnect monitoring system 200 according to another embodiment of the present invention. To avoid ambiguity, the cross-wafer interconnect monitoring system in FIG. 2 is referred to as the cross-wafer interconnect monitoring system 200. The cross-wafer interconnect monitoring system 200 includes a first wafer 10, a second wafer 11, an intra-wafer detector 10a, a first inter-wafer detector 10b, a second inter-wafer detector 10c, a neural network 12, and an intermediary 13. The intra-wafer detector 10a is embedded in the first wafer 10. The first inter-wafer detector 10b is allocated from the first wafer 10 to the second wafer 11. The second inter-wafer detector 10c is allocated from the first wafer 10 to the intermediary 13. Here, the second inter-wafer detector 10c includes a second detection circuit 10c1 and a second metal wiring interconnect layer 10c2. The second detection circuit 10c1 is embedded in the first wafer 10. The second metal wiring interconnect layer 10c2 is allocated from the first wafer 10 to the intermediary 13 as a second metal loop circuit. Here, the intermediary 13 is arranged between the first wafer 10 and the second wafer 11. The neural network 12 is connected to the intra-wafer detector 10a, the first inter-wafer detector 10b, and the second inter-wafer detector 10c. In the cross-wafer interconnect monitoring system 200, the second inter-wafer detector 10c is assigned from the first wafer 10 to the mediator 13 for detecting the third feature between the first wafer 10 and the mediator 13. The mediator 13 can be used to connect the first wafer 10 and the second wafer 11. As described above, the neural network 12 can compare the first feature with the second feature to generate the cross-wafer interconnect profile from the first wafer 10 to the second wafer 11. Similarly, the neural network 12 can compare the first feature with the third feature to generate the interconnect profile from the first wafer 10 to the mediator 13. It should be understood that the cross-wafer interconnect profile can be generated by using a bidirectional detection mechanism. For example, the intra-wafer detector 10a can be embedded in the second wafer 11. The first inter-wafer detector 10b can be assigned from the second wafer 11 to the first wafer 10. Therefore, the neural network 12 can infer the cross-wafer interconnect profile from the second wafer 11 to the first wafer 10. In addition, the cross-wafer interconnect monitoring system 100 or 200 is not limited to detecting paired wafer interconnect data. For example, when N first inter-wafer detectors 10b are introduced, the neural network 12 can infer N cross-wafer interconnect data from the first wafer 10 to the N second wafers 11. N can be any positive integer. Any reasonable technical or hardware modification belongs to the scope of the present invention.

[0013] FIG. 3 is an illustration of the neural network 12 of the cross-wafer interconnect monitoring system 100 inferring cross-wafer interconnect data. The neural network 12 may receive the first feature obtained by the intra-wafer detector 10a and the second feature obtained by the first intra-wafer detector 10b. In addition, the neural network 12 may receive the test line data of the first wafer 10 and the second wafer 11. The test line data of the first wafer 10 and the second wafer 11 may be used to assist the neural network 12 in generating cross-wafer interconnect data. In this embodiment, the test line data of the first wafer 10 and the second wafer 11 may be obtained from a wafer probe (CP) stage node, a final test (FT) stage node, or a wafer acceptance test (WAT) stage node. When the training model of the neural network 12 is fully trained, the neural network 12 may use the training model to infer cross-wafer interconnect data accordingly. It should be understood that the neural network 12 may perform an online intervention stage or an offline intervention stage, depending on its computing power. After generating cross-wafer interconnect data for all wafers of the wafer, the neural network 12 may generate a cross-wafer interconnect distribution of a wafer map. For example, the neural network 12 may generate a cross-wafer interconnect distribution of resistance-capacitance (RC) data of a wafer map. Therefore, the neural network 12 can select at least two chips from the wafer map and generate a multi-chip package based on the cross-chip interconnect distribution.

[0014] FIG. 4 is a flow chart of executing the cross-wafer interconnect monitoring method, which is performed by the cross-wafer interconnect monitoring system 100. The cross-wafer interconnect monitoring method may include steps S401 to S404. Any hardware or technical modifications fall within the scope of the present invention. Steps S401 to S404 are as follows.

[0015] Step S401: Provide a first wafer 10 and a second wafer 11.

[0016] Step S402: embedding an intra-wafer detector 10a into the first wafer 10 to detect a first feature of the first wafer 10.

[0017] Step S403: Assigning a wafer detector 10b from the first wafer 10 to the second wafer 11 to detect a second feature between the first wafer 10 and the second wafer 11;

[0018] Step S404: Compare the first feature with the second feature through the neural network 12 to generate cross-wafer interconnection data from the first chip 10 to the second chip 11.

[0019] The details of step S401 to step S404 have been previously described. Therefore, they are omitted here. In the inter-wafer interconnection monitoring system 100, the inter-wafer detector and the intra-wafer detector can be used to detect the inter-wafer interconnection patterns of all combinations. Since the inter-wafer interconnection data can be obtained through the neural network 12, an appropriate wafer combination can be selected to generate a multi-wafer package. Since the resistance-capacitance interconnection data of the selected wafer combination meets the previously defined resistance-capacitance requirements, the inter-wafer interconnection monitoring system 100 can improve the operating performance of the multi-wafer package.

[0020] In summary, the present invention discloses a cross-wafer interconnection monitoring method and a cross-wafer interconnection monitoring system. The cross-wafer interconnection monitoring system includes at least one inter-wafer detector and at least one intra-wafer detector, which are used to generate cross-wafer interconnection data from one wafer to at least one other wafer. In addition, the cross-wafer interconnection monitoring system generates a cross-wafer interconnection distribution of a wafer map. The cross-wafer interconnection distribution can directly point out an unsuitable wafer combination when a multi-wafer package fails to achieve expected performance. In addition, the cross-wafer interconnection monitoring system can select at least two wafers from the wafer map and generate a multi-wafer package with expected performance according to the cross-wafer interconnection distribution.

[0021] Those skilled in the art will quickly observe that many modifications and variations can be made to the apparatus and methods while retaining the teachings of the invention. Therefore, the above disclosure should be limited only within the scope and metes and bounds of the appended claims. The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

[0022] 10: First chip 10a: On-wafer detector 10b: First inter-wafer detector 10b1: First detection circuit 10b2: First metal wiring interconnection layer 10c: Second inter-wafer detector 10c1: Second detection circuit 10c2: Second metal wiring interconnection layer 100: Cross-wafer interconnect monitoring system 11: Second chip 12: Neural Networks 13:Mediator 200: Cross-wafer interconnect monitoring system S401: Steps S402: Steps S403: Step S404: Steps

Claims

1. A method for monitoring cross-chip interconnects includes: Provide the first chip and the second chip; An in-wafer detector is embedded in the first wafer to detect a first feature of the first wafer; an inter-wafer detector is assigned from the first wafer to the second wafer to detect a second feature between the first wafer and the second wafer; and the first feature is compared with the second feature by a neural network to generate cross-wafer interconnect data from the first wafer to the second wafer.

2. The method of claim 1, wherein the in-wafer detector and the first inter-wafer detector are ring oscillator detectors, and the first feature and the second feature include a frequency signal.

3. The method of claim 1, wherein the cross-chip interconnect data from the first chip to the second chip is a resistor-capacitor (RC) interconnect data from the first chip to the second chip.

4. The method of claim 1, wherein the first inter-wafer detector includes a first detection circuit and a first metal interconnect layer, the first detection circuit being embedded in the first wafer and the first metal interconnect layer being configured as a first metal loop circuit from the first wafer to the second wafer.

5. The method as described in claim 4, further comprising: Adjust the length of the first metal wiring interconnect layer; And receive and process at least one additional second feature between the first wafer and the second wafer.

6. The method as described in claim 1, further comprising: A second inter-wafer detector is assigned from the first wafer to an intermediary to detect a third feature between the first wafer and the intermediary; The intermediary is disposed between the first chip and the second chip and is used to connect the first chip and the second chip.

7. The method as described in claim 6, further comprising: The first feature is compared with the third feature to generate interconnect data from the first chip to the intermediary via the neural network.

8. The method as described in claim 1, further comprising: The test line data of the first chip and the second chip are obtained; and the test line data of the first chip and the second chip are input into the neural network to assist the neural network in generating the cross-chip interconnect data.

9. The method of claim 8, wherein the test line data of the first wafer and the second wafer is the test line data of the first wafer and the second wafer obtained from a wafer probe (CP) stage node, a final test (FT) stage node or a wafer acceptance test (WAT) stage node.

10. The method as described in claim 1, further comprising: This neural network generates a cross-wafer interconnect distribution of a wafer map; And select at least two wafers from the wafer diagram to generate a multi-wafer package based on the cross-wafer interconnect distribution.

11. A cross-chip interconnect monitoring system includes: First chip; Second chip; An in-wafer detector embedded in the first wafer; a first inter-wafer detector distributed from the first wafer to the second wafer; The neural network is connected to the in-wafer detector and the first inter-wafer detector; wherein the in-wafer detector is used to detect a first feature of the first wafer, the first inter-wafer detector is used to detect a second feature between the first wafer and the second wafer, and the neural network compares the first feature with the second feature to generate cross-wafer interconnect data from the first wafer to the second wafer.

12. The system of claim 11, wherein the in-wafer detector and the first inter-wafer detector are ring oscillator detectors, and the first feature and the second feature include a frequency signal.

13. The system of claim 11, wherein the cross-chip interconnect data from the first chip to the second chip is resistor-capacitor (RC) interconnect data from the first chip to the second chip.

14. The system of claim 11, wherein the first cross-wafer detector includes a first detection circuit and a first metal interconnect layer, the first detection circuit being embedded in the first wafer, and the first metal interconnect layer being configured as a metal loop circuit from the first wafer to the second wafer.

15. The system of claim 14, wherein the length of the first metal interconnect layer is adjustable, and the neural network receives and processes at least one additional second feature between the first wafer and the second wafer.

16. The system as described in claim 11, further comprising: A second cross-chip detector; And an intermediary layer placed between the first wafer and the second wafer; The second cross-wafer detector is configured to detect a third feature between the first wafer and the interposer layer, and the interposer layer is used to connect the first wafer and the second wafer.

17. The system of claim 16, wherein the neural network compares the first feature with the third feature to generate interconnect data from the first wafer to the interposer.

18. The system of claim 11, wherein the neural network acquires test line data of the first wafer and the second wafer, and the test line data of the first wafer and the second wafer is used to assist the neural network in generating the cross-wafer interconnect data.

19. The system of claim 18, wherein the test line data of the first wafer and the second wafer is obtained from a wafer probe (CP) stage node, a final test (FT) stage node, or a wafer acceptance test (WAT) stage node.

20. The system of claim 11, wherein the neural network generates a cross-wafer interconnect distribution of a wafer map, and the neural network selects at least two wafers from the wafer map to generate a multi-wafer package based on the cross-wafer interconnect distribution.

Citation Information

Patent Citations

  • Semiconductor timing improvement

    US20140143748A1

  • Fingerprinting of semiconductor die arrangements

    US20220179950A1

  • Apparatus and method for verifying process integrity

    US6463570B1