Method of using partially defective programmable logic devices

[0002] The present invention relates to programmable logic devices, and more particularly to a method for using programmable logic devices that contain minor defects.

[0003] A programmable logic device, such as a field programmable gate array (FPGA), is designed to be user-programmable so that users can implement logic designs of their choices. In a typical architecture, an FPGA includes an array of configurable logic blocks (CLBs) surrounded by programmable input/output blocks (IOBs). The CLBs and IOBs are interconnected by a hierarchy of programmable routing resources. These CLBs, IOBs, and programmable routing resources are customized by loading a configuration bitstream into configuration memory cells of the FPGA. Additional resources, such as multipliers and memory, may be included.

[0004] There are strong customer demands for FPGAs that have a large number of CLBs, IOBs, and/or other resources (e.g., multipliers and block RAMs). This is because end products are becoming more complex, which require more CLBs and IOBs to implement complicated designs. As a result, the size of FPGA die and the resources fabricated on the die grow. This means that the chance of finding a defect in a die increases because the number of defects is proportional to the area of a die and the complexity of technology.

[0005] The circuits implemented by different customers of FPGAs are unique. Further, a circuit design may undergo modifications during product development phase. Thus, multiple versions of a circuit may be implemented on a FPGA. In other words, FPGAs are not design specific because they can be theoretically used in any design. This situation places a heavy burden on the quality and reliability of the FPGAs. If a FPGA contains a single defect (e.g., one of its configuration memory cells is defective), it may render an end product unusable because the design may need to use that defective resource. In order to avoid problems with customers, a FPGA manufacturer needs to discard a FPGA even if it contains only one defect.

[0006] The problem of low yield has significant economic impact on FPGA manufacturers. There are two types of defects: gross defect (that causes failure of an entire FPGA) and localized defect (that causes failure of small circuitry in the FPGA). It has been found that close to two thirds of large FPGA dies are discarded because of localized defects. If a method can be found to use some of these defective dies, the cost of product of the FPGA manufacturer could be reduced significantly. As a result, customers can take advantage of lower priced FPGAs for specific design patterns.

[0007] The present invention is a method for using a FPGA that contains at least one localized defect. A design is loaded into the FPGA. The FPGA is tested to determine whether it can execute the design accurately even with the localized defect. For example, if the design does not use the localized defect, the localized defect would not affect the execution of the design. In this case, the FPGA is accepted as suitable for this design. Even if the FPGA is found to be unsuitable for a specific design pattern, it may still be suitable for other designs. Thus, in another embodiment of the present invention, additional loading and testing of designs are performed.

[0008] By using this method, a FPGA manufacturer can sell FPGAs that are normally discarded. As a result, the price of these FPGAs could be very low.

[0009] The present invention is useful for FPGA customers that have finalized their designs. At that time, the design is fixed, and the above mentioned method can be used to determine whether a FPGA with localized defects can be used to implement the design. This method is especially useful for FGPA customers that are considering whether to convert a finalized design from FPGA to an application specific integrated circuit (ASIC). The FPGAs selected in accordance with the present invention can be very price competitive with ASICs. Further, no conversion from one type of device (FPGA) to another type (ASIC) is needed. This means that the “customer specific” integrated circuits are timing and functionality equivalent to the integrated circuits used in product development phase. Thus, it opens another possibility to the customers.

[0010] The above summary of the present invention is not intended to describe each disclosed embodiment of the present invention. The figures and detailed description that follow provide additional example embodiments and aspects of the present invention.

[0011] FIG. 1 is a flow chart showing the operation of the present invention.

[0012] FIG. 2 is a schematic diagram showing an FPGA having a localized defect and the relationship of several designs to the localized defect.

[0013] FIG. 3 is a flow chart showing another embodiment of the present invention.

[0014] FIG. 4 is a schematic diagram showing die in a wafer and an associated file.

[0015] The present invention relates to programmable logic devices. In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced without these specific details. In other instances, well-known features have not been described in detail in order to avoid obscuring the present invention.

[0016] FIG. 1 is a flow chart showing one embodiment of the present invention as applied to FPGAs. In step 102, fabricated dies in wafers are set up for testing. In step 104, gross defects are tested. This is typically a DC type of testing. Examples of gross defects are opens, shorts, and excessive leakage currents. If gross defects are found, the die is rejected (step 106). If there are no gross defects, test for localized defects follows. The memory cells are first tested (step 108). This test can be performed by writing digital data to the cells and reading them back. If there is no localized defect in the memory cells, the basic functions of the FPGA are tested (step 110). If there is no localized defect in the basic functions, performance testing is performed (step 112). An example of a performance test is speed performance testing. It is known that different integrated circuits from different fab lots or wafers may have different speed performance due to process variation. In order to make sure that a FPGA meets its manufacturer's speed performance specification, the FPGA can be configured to have different paths and then measure the time ‘delay’ of signal travel through each path. Usually, the FPGA that has less delay can handle circuits that need high speed performance. If the performance testing is passed, the FPGA is not defective, and is placed in a specially designated bin (step 114). If performance testing indicates that the FPGA does not meet the minimum performance specification, the FPGA is rejected (step 116).

[0017] The cases where the FPGA contains localized defects in memory cells (i.e., does not pass step 108) or basic functions (i.e., does not pass step 110) is now described. In prior art procedures, the FPGA is discarded. However, in the present invention, the FPGA is further tested to determine if it may be used to implement a specific customer design (step 120). In performing this step, the customer's design is loaded into the FPGA and the FPGA is configured. The FPGA is tested to make sure that the design works within specification. If the test reveals that the design is adversely affected by the localized defects in the FPGA, the FPGA is rejected (step 122). If the FPGA passes the test, a performance test is performed (step 124). If the FPGA fails to meet the minimum performance specification, it is again rejected. If it passes performance testing, the FPGA is placed in another specially designated bin (step 126). The FPGAs in this bin can only be used to implement this specific customer design.

[0018] The above described procedure works because many customer designs use a small portion (e.g., less than 10% to 15%) of the total FPGA available resources. For example, a FPGA contains many interconnect lines to allow users flexibility in designing their circuits. Many of these interconnect lines are not used in a specific design. Thus, if a FPGA contains only a few localized defects, it is likely that many designs do not encounter these localized defects. As a result, the FPGA is usable for that particular design. There is no need to discard the FPGA.

[0019] As an improvement to the embodiment of FIG. 1, several customer designs can be tested. In case the FPGA is not useable for one design because this design encounters a localized defect, it is possible that the same FPGA may be used to implement another design. For example, FIG. 2 shows a FPGA 200 having one localized defect (shown as an “X” inside FPGA 200). Design A needs to use resources that include the localized defect, and thus FPGA 200 is not suitable to implement design A. However, design B does not involve this localized defect, and FPGA 200 may be used to implement design B. The same principle applies if FPGA 200 contains many localized defects.

[0020] FIG. 3 shows the testing performed after steps 108 and 110 indicates that the FPGA contains localized defects (shown in FIGS. 1 and 3 as line 118). Steps 120, 122, 124 and 126 in FIG. 3 are the same as the steps of the same reference numerals in FIG. 1. After the design in step 120 is found to be unsuitable for this FPGA, the FPGA is then tested for design B (step 140). If it is found that this FPGA is not suitable to implement design B, the next design is tested. If it is determined that this FPGA can be used to implement design B, a performance testing is performed (step 144). If the FPGA fails the performance testing, it is again rejected. If it passes the performance testing, the FPGA is placed in another specially designated bin (step 146). The FPGAs in this bin can only be used to implement this specific customer design.

[0021] The above described procedure is used until the last design (shown in FIG. 3 as design M) is tested.

[0022] The application of the above described procedure in a wafer is now described. A wafer typically contains a number of dies (such as the dies marked “1”, “2”, . . . , “32”, in a wafer 212 of FIG. 4). All, some, or none of the dies may contain localized defects. In one embodiment, each die is tested using the procedures described in FIG. 1 or 3. The result of the testing is stored in an electronic file 214. Each record in the file contains the identification of the corresponding die (i.e., wafer number in column 1 of file 214 and location of the die on the wafer in column 2 of file 214) and the design number assigned to this die. In FIG. 4, a non-defective die has a design number of “GD,” (or other symbols) while dies having localized defects are associated with design numbers “A,” “B,” “C,” etc. This file is used during assembly. In some cases, a customer wants to use different packages for different designs. In these cases, the file also associates a die with a package, depending on which customer design is associated with the die. One advantage of this method is that there is no need to apply ink to mark the die (as is done in some conventional system). After assembly, the finished package of a die may be marked with a special code to indicate that this FPGA is certified for a specific purpose. This code is visible to end users so as to avoid incorrectly using the FPGA. Prior to shipping the package, it may optionally go through additional testing to make sure that nothing is damaged during the assembly process.

[0023] It can be seen from the above description that a novel method to use defective FPGAs has been disclosed. Those having skill in the relevant arts of the invention will now perceive various modifications and additions which may be made as a result of the disclosure herein. Accordingly, all such modifications and additions are deemed to be within the scope of the invention, which is to be limited only by the appended claims and their equivalents.