Method and system for obtaining a specification limit value for use as acceptance criterion in a device acceptance test

Abstract

A method of obtaining at least one of an upper and a lower specification limit value for use as acceptance criterion in a device acceptance test, wherein the upper specification limit value (USL) is obtainable by adding a deviation to a central value (X50) and the lower specification limit value (LSI) is obtainable by subtracting the deviation from the central value (X50), wherein the deviation is at least based on a spread value (s), comprises obtaining (13,15,16) a set of parametric test result values, each obtained by carrying out the parametric test on a respective one of a plurality of test devices (2). The method further comprises obtaining the central value (X50) and determining, from the parametric test result values, the deviation. Determining the deviation comprises determining a first parametric test result value (X0.621) corresponding to a percentile function value at a first percentile of the set of parametric test result values and determining a second parametric test result value (X99.379) corresponding to a percentile function value at a second percentile of the set of parametric test result values, wherein the first percentile is below 50 % and the second percentile is above 50 %. The spread value (s) is determined on the basis of both the first parametric test result value (X0.621) and the second parametric test result value (X99.379).

Description

[0001] M / ADVA-017-PC 1

[0002] METHOD AND SYSTEM FOR OBTAINING A SPECIFICATION LIMIT VALUE FOR USE AS ACCEPTANCE CRITERION IN A DEVICE ACCEPTANCE TEST

[0003] The invention relates to a method of obtaining at least one of an upper and a lower specification limit value for use as acceptance criterion in a device acceptance test,

[0004] wherein the upper specification limit value is obtainable by adding a deviation to a central value and

[0005] the lower specification limit value is obtainable by subtracting the deviation from the central value,

[0006] wherein the deviation is at least based on a spread value, the method comprising:

[0007] obtaining a set of parametric test result values, each obtained by carrying out the parametric test on a respective one of a plurality of test devices;

[0008] obtaining the central value;

[0009] determining, from the parametric test result values, the deviation.

[0010] The invention also relates to a system for generating at least one of an upper and a lower specification limit value for use as acceptance criterion in a device acceptance test.

[0011] The invention also relates to a computer program.

[0012] WO 2023 / 137312 Al discloses a method of determining an anomaly in a manufacturing process. The method includes receiving series data from a piece of manufacturing or fabrication, metrology or inspection equipment. The method includes segmenting the time-series data and classifying and labelling the segments as being associated with a transient state part of the signal and a steady-state part of the signal. Following the classification and state labelling, the method includes performing a set of dynamic-based analyses and static- M / ADVA-017-PC 2

[0013] based analyses. A list of analytical features that can be employed for a feature assessment module includes a count of the number of pulses at steady state, corresponding to a count of the number of excursions from an acceptable range (2σT) around a steady-state level.

[0014] More generally, developers of test programs for use in device acceptance tests, tests that result in a pass or a fail, specify an upper and a lower specification limit to be applied to the result of a parametric test, a test resulting in a value of a quantity. The upper specification limit value is obtained by adding a deviation to a mean value and the lower specification limit value is obtained by subtracting a deviation from the mean value, with the deviation usually being a multiple of a standard deviation determined for a set of test parametric test result values, each obtained by carrying out the parametric test on a respective one of a plurality of test devices. However, this only results in acceptable yields and accurate discrimination between good and bad devices if the test values have a normal distribution about the value for a good device. In practice, there may be a typical device fault resulting in a typical parametric test result value different from that typical for good devices. The parametric test result values for bad devices will have a certain distribution around this typical deviating value for devices having the fault. Setting the range of acceptable values from the lower specification limit value to the higher specification limit value as a range centred on the parametric rest result value typical for a good device (i.e. a device not having the fault) will result either in bad devices being classified as good, if the range is too large, or relatively low yields, if the range is small.

[0015] It is an object of the invention to provide a method, system and computer program allowing a designer of a test program to obtain at least one of an upper and a lower specification limit value that are more useful as acceptance criterion in a device acceptance test.

[0016] The problem is solved according to a first aspect by the method according to the invention, which method comprises: M / ADVA-017-PC 3

[0017] obtaining a set of parametric test result values, each obtained by carrying out the parametric test on a respective one of a plurality of test devices;

[0018] obtaining the central value;

[0019] determining, from the parametric test result values, the deviation, wherein determining the deviation comprises

[0020] determining a first parametric test result value corresponding to a percentile function value at a first percentile of the set of parametric test result values and

[0021] determining a second parametric test result value corresponding to a percentile function value at a second percentile of the set of parametric test result values,

[0022] wherein the first percentile is below 50 % and the second percentile is above 50 %; and

[0023] determining the spread value on the basis of both the first parametric test result value and the second parametric test result value.

[0024] The method will generally be implemented on a computer. Because the method comprises obtaining a set of parametric test result values, each obtained by carrying out the parametric test on a respective one of a plurality of test devices, the upper and lower specification limit values represent achievable tolerances. Thus, sampling error and variance in the production itself are taken into account. It follows that using this set of parametric test result values is a factor in ensuring that the device yield that is obtained when the acceptance test is implemented in a production environment is not too low. Because the method comprises determining a first parametric test result value corresponding to a percentile function value at a first percentile of the set of parametric test result values and determining a second parametric test result value corresponding to a percentile function value at a second percentile of the set of parametric test result values, wherein the first percentile is lower than 50 % and the second percentile is higher than 50 %, and because the spread value is determined on the basis of both the first and the second parametric test result value, the method is suitable for parametric test values that do not exhibit a normal distribution. If there is a large range of values to one M / ADVA-017-PC 4

[0025] side of the central value, then the spread value, and thus the deviation, will be larger. This is the case regardless of whether this large range is a range of values lower than the central value or a range of values higher than the central value.

[0026] The method may be used to obtain only one of the upper and lower specification limit values that are actually used as acceptance criterion in the device acceptance test. The other limit value may follow from some other constraint. For example, in the case of noise measurement, the lower limit value may correspond to the lowest measurable value. The upper limit value may then be calculated using the method.

[0027] The method is not limited to obtaining acceptance criteria for a device acceptance test for a particular type of device. Thus, a device may be a singu-lated or partially assembled die, a wafer, a printed circuit board or a high-level assembly (an assembly of different modules, e.g. different printed circuit boards). The device may also be a memory chip.

[0028] The central value may correspond to the mean value, so that obtaining the central value may comprise calculating the mean value of the set of parametric test result values. The central value may alternatively be a target value set by a designer of the device, so that obtaining the central value may comprise no more than obtaining the central value as an input, e.g. through a user interface or data exchange interface.

[0029] The deviation is at least based on a spread value, being either based on the spread value or corresponding to the spread value. Determining a parametric test result value corresponding to a percentile function value at a certain first percentile of the set of parametric test result values does not need to entail obtaining the percentile function as a mathematical formula. It is sufficient to determine the value below which the certain percentile of values in the set lie.

[0030] In an embodiment, the spread value (s) is determined as a maximum of: M / ADVA-017-PC 5

[0031] (i) a difference between the second parametric test result value and the central value and

[0032] (ii) a difference between the central value

[0033]

[0034] and the first parametric test result value.

[0035] This embodiment is particularly suited to parametric tests that yield asymmetric parametric test result value distributions. There may be a long tail of values even for devices that should fall in the class of devices that the acceptance test is intended to select, i.e. devices that should pass the test.

[0036] In an embodiment of the method, the deviation is proportional to the spread value.

[0037] The proportionality factor will be different from a fixed value of one. Such a factor can be adjusted to take account of other restrictions or requirements than a requirement to capture most of the distribution of parametric test result values associated with devices that should pass the acceptance test.

[0038] In a particular variant of this embodiment, the deviation is obtained by multiplying the spread value (s) by a proportionality factor dependent on at least one of:

[0039] the number of values in the set;

[0040] a target confidence level; and

[0041] a target yield.

[0042] Where the proportionality factor is dependent on the number of values in the set, it is possible to narrow down the deviation with increasing number of values in the set. This reflects an increase in confidence that the values at the edge of the distribution are associated with devices that pass the test. Similar effects can be achieved where the proportionality factor is dependent on a target confidence level. A lower target confidence level may result in a smaller interval between the upper specification limit value and the lower specification limit value, reflecting that it is less certain that the target yield will be met. M / ADVA-017-PC 6

[0043] There will be more false negatives for a smaller interval. If the target yield is higher, the proportionality factor can be increased, to ensure that the targeted proportion of devices as classified as a "pass".

[0044] Thus, in an example of any embodiment of the method in which the deviation is proportional to the spread value, the deviation is obtained by multiplying the spread value by a proportionality factor that is a function of the number of values in the set and converges towards a boundary value with increasing number of values in the set.

[0045] The boundary value ensures that a target yield can still be met.

[0046] An embodiment of the method comprises performing regression analysis on the set of parametric test result values to determine an expected yield when applying the upper specification limit value and the lower specification value for a given confidence interval.

[0047] A designer of the device or the parametric and acceptance tests can thus assess whether large-scale production of the devices using the acceptance test would be economical. If it would not be, the design of one or both of the tests or of the device can be adapted.

[0048] An example of any embodiment of the method in which the deviation is proportional to the spread value and the method comprises performing regression analysis on the set of parametric test result values to determine an expected yield when applying the upper specification limit value and the lower specification value for a given confidence interval further comprises adjusting the proportionality factor in dependence on a comparison of the expected yield with the target yield.

[0049] Thus, if the device design allows, the acceptance test is adjusted to achieve a target yield. The embodiment allows to achieve the target yield without needing to modify either the device design or the design of the parametric test. M / ADVA-017-PC 7

[0050] An example of any embodiment that comprises performing regression analysis on the set of parametric test result values to determine an expected yield when applying the upper specification limit value and the lower specification value for a given confidence interval further comprises comparing the expected yield with a target yield, and, if the expected yield is lower than the target yield, carrying out at least one of

[0051] a step of adjusting the target yield and

[0052] a step comprising modifying a device design, manufacturing a number of prototype devices according to the modified device design, carrying out the parametric test on the prototype devices to obtain a new set of parametric test result values and applying the method to the new set of parametric test result values.

[0053] In this embodiment, at least one of the acceptance test design and the device design is adapted to be able to guarantee with a certain degree of confidence a particular yield.

[0054] An embodiment of the method comprises

[0055] obtaining a target specification,

[0056] determining whether the upper specification limit value and the lower specification limit value are more restrictive on an allowable range of parametric test result values than the target specification and,

[0057] if the upper specification limit value and the lower specification limit value are not more restrictive on an allowable range of parametric test result values than the target specification, carrying out at least one of

[0058] a step of adjusting the target specification,

[0059] a step comprising modifying a device design, manufacturing a number of prototype devices according to the modified device design, carrying out the parametric test on the prototype devices to obtain a new set of parametric test result values and applying the method to the new set of parametric test result values, and

[0060] a step of adjusting a target yield. M / ADVA-017-PC 8

[0061] In this embodiment, the upper and lower specification limit values obtained using the method are only used in the parametric test if they are more restrictive than the target specification. The target specification comprises equivalent bounds to the range of acceptable parametric test result values. In this embodiment, it is ensured the range resulting from the method is more restrictive. In a sense, the method is used to determine whether the device design and the design of the parametric test are capable of meeting the target specification in practice, given in particular the variability of test results.

[0062] In an embodiment of the method, obtaining the set of parametric test result values comprises

[0063] obtaining a raw data set of parametric test result values,

[0064] applying at least one anomaly detection technique to the raw data set, and

[0065] removing parametric test result values determined to be anomalies from the raw data set.

[0066] It is noted that any embodiment that additionally comprises determining an expected yield when applying the upper specification limit value and the lower specification value for a given confidence interval may use the raw data set to determine the expected yield. Using the set of values with the anomalies removed to calculate the upper and lower specification limit values results in a smaller interval between the upper and lower specification limit values.

[0067] An embodiment of the method further comprises deploying the obtained ones of the upper specification limit value and the lower specification limit value to automated test equipment for use in a production environment for producing the devices.

[0068] An example of this embodiment further comprises carrying out parametric testing and an acceptance test using the deployed specification limit values on devices produced in the production environment. M / ADVA-017-PC 9

[0069] According to another aspect, the system according to the invention comprises:

[0070] an interface for obtaining parametric test result values;

[0071] an interface for returning the generated specification limit values; and a data processing system, configured to carry out a method according to the invention.

[0072] According to a further aspect of the invention, there is provided a computer program comprising instructions which, when the program is executed by a computer comprising an interface for obtaining parametric test result values and an interface for returning specification limit values, cause the computer to carry out a method according to the invention.

[0073] According to yet a further aspect of the invention, there is provided a computer-readable medium having stored thereon a computer program according to the invention.

[0074] The invention will be explained in further detail with reference to the accompanying drawings, in which:

[0075] Fig. 1 is a schematic diagram showing a test cell and a system for generating upper and lower specification limit values for use as acceptance criterion in a device acceptance test carried out in the test cell;

[0076] Fig. 2 is a histogram of parametric test result values illustrating the effect of the choice of upper and lower specification limit values on the effectiveness of the device acceptance test;

[0077] Fig. 3 is a quantile plot of raw parametric test result values and a curve fit to a sub-set thereof, including a confidence band; and

[0078] Fig. 4 is a flow chart of a method of generating the upper and lower specification limit values.

[0079] A test cell 1 in an environment for producing devices 2 comprises a test cell controller 3, testing apparatus 4 and material handling equipment 5. The material handling equipment 5 comprises a prober 6 and a handler 7. M / ADVA-017-PC 10

[0080] The device 2 may be a semiconductor device, e.g. a semiconductor wafer, one of an array of devices at least partially formed on the semiconductor wafer, e.g. prior to singulation, a system on a chip, an integrated circuit, Printed Circuit Board (PCB), etc.

[0081] The prober 6 may comprise a socket or probe card comprising a plurality of contact pins for engaging the device 2. The handler 7 is arranged to position the device 2 for testing. The handler 7 of the illustrated test cell 1 is also arranged to control the testing apparatus 4, in that the handler 7 synchronises the device placement on the prober 6 with the start of the testing executed by the testing apparatus 4. The testing apparatus 4 is arranged to obtain raw data in the form of parametric test result values. For example, the device 2 may be provided with a test signal and a response signal obtained from the device 2. The parametric test result value may comprise the value of the response signal or of a characteristic of the response signal. Obtaining the raw data may comprise processing one or more response signals by means of one or more analogue or digital filters.

[0082] The raw data is the basis for an acceptance test indicating whether the device 2 meets acceptance criteria or not. The acceptance test may simply discriminate between good and bad devices, the bad devices becoming rejects. Alternatively, the acceptance test may discriminate between devices of different quality classes associated with respective fields of application. A designer of the device 2 or of an acceptance test program for the device 2 will need to design the device 2 and the parametric test, as well as specifying at least one of an upper specification limit value USL and a lower specification limit value LSL (Fig. 3) for the parameter of which the value is determined in the parametric test.

[0083] There is a first distribution 8 (Fig. 2) of parametric test result values for acceptable devices 2 and a second distribution 9 of parametric test result values for unacceptable devices 2 due to e.g. sampling error or variations in the production process of the device 2. The range covered by the LSL and USL M / ADVA-017-PC 11

[0084] (shaded in Fig. 2) should be such as to encompass as large a proportion of the acceptable devices 2 as possible, this being the yield, whilst excluding the unacceptable devices 2. In this regard, it is noted that the first distribution 8 is a non-normal distribution. In the illustrated example, the first distribution 8 exhibits both skew and a high degree of kurtosis ("tailedness").

[0085] A method of obtaining the USL and LSL values (Fig. 4) using a set of prototype device 2 is carried out by a data processing system 10 (Fig. 1) comprising a network interface 11 to a network 12. The method comprises a first step 13 of performing the parametric test on the prototype devices 2 to obtain a raw data set of parametric test result values. This step 13 may comprise the test cell controller 3 obtaining the raw data from the testing apparatus 4 and making this data available to the data processing system 10 via a network interface 14 and the network 12.

[0086] In a next step 15, a data quality check is carried out. This step 15 may be partly or completely carried out by a human.

[0087] In a subsequent step 16, the raw data set is analysed to detect outliers 17 (Fig. 3) or more generally anomalies, which are removed. The result is a set of parametric test result values for use in the remainder of the method. This step 16 may comprise the execution of at least one of a parametric-based method, a network-based method, a clustering method, an ensemble technique and classification using a neural network. As an example, the GESD (generalised extreme studentised deviate) test, a random forest method, a method comprising determining interquartile ranges (IQR) or a Hampel filter may be used.

[0088] A next step 18 comprises calculating the USL and LSL. The calculation is performed according to the following formula:

[0089] USL, LSL = central value ± k·s. (1) M / ADVA-017-PC 12

[0090] Here, the spread s is a value based on the distribution of parametric test result values in the set obtained in the preceding step 16. More particularly, determining a value of the spread s comprises determining a first parametric test result value corresponding to a first percentile function value at a first percentile of the set of parametric test result values, the first percentile being below 50 %. In other words, the (first) parametric test result value for which a first percentile of the set of parametric test result values is smaller than that first parametric test result value is determined. Determining a value of the spread salso comprises determining a second parametric test result value corresponding to a second percentile function value at a second percentile of the set of parametric test result values, the second percentile being above 50 %. In other words, the (second) parametric test result value for which a second percentile of the set of parametric test result values is smaller than that second parametric test result value is determined. Thus, measures of the boundaries of the distribution of parametric test result values in the set are obtained. The spread s is determined as either a difference between the second parametric test result value and the central value or a difference between the central value and the first parametric test result value, whichever of the two is larger.

[0091] The central value taken here as an example corresponds to a median value x50, but may alternatively be the mean or a target value specified by the designer of the device 2.

[0092] Thus, in an example, the upper and lower specification limit values USL, LSL are obtained using the following formula:

[0093] USL,LSL = x50± k·max[x99.379− x50, x50− x0.621] (2)

[0094] The proportionality factor k depends on at least one of: the type of distribution of the data; the number of values in the set; a confidence level, a target yield and whether the device acceptance test requires one-sided or two-sided M / ADVA-017-PC 13

[0095] limits. The type of distribution may be determined by fitting the values to different probability density functions. Alternatively, a default type may be used, e.g. Gaussian, in which case the proportionality factor k does not depend on the type of distribution. The confidence level is user-defined, e.g. 95 %.

[0096] In particular where k does not depend exclusively on the number of values, the proportionality factor k may be the product of multiple proportionality factors, one of which is a function of the type described.

[0097] A suitable formula for a two-sided normal distribution is:

[0098] k = w · k2, (3)

[0099] k2= z1+p / 2· √(v·(1+1 / N) / χ²1-α,v) (4) %l-a,v

[0100] χ²N-1,1-α

[0101] (5)

[0102]

[0103] 2·(N+1)2

[0104] Here, N is the number of values, p is the target yield, a is the confidence level, v is the number of degrees of freedom, in this example equal to N-l, z is the critical value of a normal distribution for cumulative probability 1 + p / 2, and χ²1-α,vis the critical value of a chi-square distribution with degrees of freedom v that is exceeded with probability α. Where the factor k also depends on a target yield, a confidence level or both, as in the illustrated example, the value may be determined or adjusted iteratively, as will be explained.

[0105] In the illustrated example, the designer will already have had values of the upper and lower specification limit values in mind, based on design considerations such as target specified by a customer. These must be met in any case. Thus, in a next step 19, the values USL, LSL calculated using the formula (2) or another version of the more general formula (1) are compared with the target specification. Only if the calculated values USL, LSL lie within a range specified M / ADVA-017-PC 14

[0106] as the target specification (e.g. in the form of a minimum and a maximum value) does the method proceed. Otherwise, the designer will perform a drill -down analysis (step 20) to investigate the cause. The designer will then adapt the design of the device 2, adjust the target specification or adjust the proportionality factor A by adjusting the target yield (step 21). A combination of these adjustments is, of course, also possible. If the design is adapted, then a new series of prototype devices 2 will need to be produced and the method returns to the very first step 13. If not, then an iteration that skips the first step 13 is performed.

[0107] As mentioned, if the calculated values USL, LSL lie within a range specified as the target specification, the method proceeds to a next step 22. This step comprises regression analysis to obtain information corresponding to a fitted curve 23 (Fig. 3) representing a linear percentile function and calculation of a confidence band 24. The analysis and calculation can be effected using a Monte Carlo method. This step 22 may be carried out on the set of parametric test result values obtained in the first step 13 or the set obtained upon performing the next two steps 15,16. The step 22 returns the expected yield for a given confidence level. The confidence level specifies quantitatively how well the parametric test result values fit the curve 23.

[0108] If the expected yield is less than a target minimum yield, then the method proceeds to the analysis and adaptation steps 20,21 discussed above.

[0109] In a variant (not shown), the upper and lower specification limit values USL, LSL are adjusted towards the central value (corresponding to a shift from the solid to the dashed lines in Fig. 3) by adjusting the value of the proportionality factor Ar, such that the yield at a target confidence level is or approximates to within a certain tolerance range a target yield.

[0110] Next, measurement system analysis is performed, in order to determine whether one or more repeatability and reproducibility criteria are met. If they are not, the method returns to the analysis step 20 introduced above. M / ADVA-017-PC 15

[0111] The measurement system analysis is a standard Gauge Repeatability and Reproducibility (R& R) study, designed to assess variation in the measurement system. The study determines whether the variation originates from the measurement system itself (repeatability) or from the production process (reproducibility). A step 25 of performing the reproducibility analysis and a step 26 of performing the repeatability analysis are similar, but differ in terms of the raw data set used for the test.

[0112] The reproducibility analysis examines process capability across both multiple test cells 1 and multiple devices 2. The number of devices 2 considered may be more than 25, with more than five devices 2 being considered for each test cell 1 considered.

[0113] The repeatability analysis considers a smaller number of devices 2, e.g. between three and five, each tested multiple times, e.g. more than 25 times, under the same configuration and environmental conditions. This ensures reliable measurement performance.

[0114] The steps 25,26 can comprise calculating Clements' Ca using interpolated quantiles of the distribution best fit to the raw data. The calculation is carried out using the following formula:

[0115] USL -x50^50 LSL Cpk = min (3)

[0116]

[0117] -x99.379-x0.621x99.379-x0.621-

[0118] The value may be reported for information purposes. In a variant, the method comprises determining whether Clements' CPk is at least equal to a pre-determined value, e.g. 1.66, and the method reverts to the analysis step 20 if the criterion is not met. Using Clements' CPk allows the method to provide good results even though the first distribution 8 is not a normal distribution.

[0119] Next or concurrently, the upper and lower specification limit values USL, LSL are deployed (step 27) to the test cell 1. They may be communicated to the M / ADVA-017-PC 16

[0120] test cell controller 3 via the network 12, for example. This may involve integrating these values in a test program with which the test cell controller 3 or a controller (not shown) of the testing apparatus 4 is programmed.

[0121] From then on, the parametric test and acceptance test are performed on pro-duction devices 2. Continuous statistical monitoring (step 28) is performed. This step 28 may comprise generating Pareto charts, control charts or both, for example.

[0122] The method is repeated whenever there are major changes of hardware, software or process (production process or parametric test process).

[0123] The invention is not limited to the embodiments explained above, which may be varied within the scope of the accompanying claims. For example, parametric test result values need not be values corresponding to or derived from a response signal from the device 2 under test. An optical test may result in a parametric test result value characterising a surface characteristic or material composition of the device 2, for example.

[0124] The step 25 of performing measurement system analysis may be modified or omitted, since the upper and lower specification limit values USL, LSL obtained prior to this step 25 will generally already deliver the effects of the invention. M / ADVA-017-PC 17

[0125] List of reference numerals

[0126] 1 - Test cell

[0127] 2 - Device

[0128] 3 - Test cell controller

[0129] 4 - Testing apparatus

[0130] 5 - Material handling equipment

[0131] 6 - Prober

[0132] 7 - Handler

[0133] 8 - First distribution

[0134] 9 - Second distribution

[0135] 10 - Data processing system

[0136] 11 - Network interface

[0137] 12 - Network

[0138] 13 - Step (perform parametric test on prototypes)

[0139] 14 -- Network interface of test cell controller

[0140] 15 - Step (check data quality)

[0141] 16 - Step (perform algorithmic outlier analysis and removal) 17 - Outlier

[0142] 18 - Step (calculate specification limits)

[0143] 19 - Step (compare with target specification)

[0144] 20 - Step (perform analysis)

[0145] 21 - Step (adapt design, specifications, expected yield) 22 - Step (calculate expected yield)

[0146] 23 - Fitted curve

[0147] 24 - Confidence band

[0148] 25 - Step (perform reproducibility analysis)

[0149] 26 - Step (perform repeatability analysis)

[0150] 27 - Step (deploy calculated limits) )

[0151] 28 - Step (perform continuous monitoring)

[0152]

Claims

M / ADVA-017-PC 18Claims1. Method of obtaining at least one of an upper and a lower specification limit value for use as acceptance criterion in a device acceptance test, wherein the upper specification limit value (USL) is obtainable by adding a deviation to a central value x50) andthe lower specification limit value (LSL) is obtainable by subtracting the deviation from the central value x50),wherein the deviation is at least based on a spread value (s), the method comprising:obtaining (13,15,16) a set of parametric test result values, each obtained by carrying out the parametric test on a respective one of a plurality of test devices (2);obtaining the central value x50,determining, from the parametric test result values, the deviation, wherein determining the deviation comprisesdetermining a first parametric test result value x0.621) corresponding to a percentile function value at a first percentile of the set of parametric test result values anddetermining a second parametric test result value x99.379) corresponding to a percentile function value at a second percentile of the set of parametric test result values,wherein the first percentile is below 50 % and the second percentile is above 50 %; anddetermining the spread value (s) on the basis of both the first parametric test result value x0.621) and the second parametric test result value x99.379).

2. Method according to claim 1,wherein the spread value (s) is determined as a maximum of: (i) a difference between the second parametric test result value X99.379) and the central value x50) andM / ADVA-017-PC 19(ii) a difference between the central value (x50) and the first parametric test result value x0.621).

3. Method according to claim 1 or 2,wherein the deviation is proportional to the spread value (s).

4. Method according to claim 3,wherein the deviation is obtained by multiplying the spread value (s) by a proportionality factor dependent on at least one of:the number of values in the set;a target confidence level; anda target yield.

5. Method according to claim 3 or 4,wherein the deviation is obtained by multiplying the spread value (s) by a proportionality factor that is a function of the number of values in the set and converges towards a boundary value with increasing number of values in the set.

6. Method according to any one of the preceding claims,performing regression analysis on the set of parametric test result values to determine an expected yield when applying the upper specification limit value (USL) and the lower specification value (LSL) for a given confidence interval.

7. Method according to any one of claims 3-5 in combination with claim 6, further comprisingadjusting the proportionality factor in dependence on a comparison of the expected yield with the target yield.

8. Method according to claim 6 or 7, further comprisingcomparing the expected yield with a target yield, and, if the expected yield is lower than the target yield, carrying out at least one ofM / ADVA-017-PC 20a step of adjusting the target yield anda step comprising modifying a device design, manufacturing a number of prototype devices (2) according to the modified device design, carrying out the parametric test on the prototype devices (2) to obtain a new set of parametric test result values and applying the method to the new set of parametric test result values.

9. Method according to any one of the preceding claims, comprising obtaining a target specification,determining whether the upper specification limit value (USL) and the lower specification limit value (LSL) are more restrictive on an allowable range of parametric test result values than the target specification and,if the upper specification limit value (USL) and the lower specification limit value (LSL) are not more restrictive on an allowable range of parametric test result values than the target specification, carrying out at least one ofa step of adjusting the target specification,a step comprising modifying a device design, manufacturing a number of prototype devices (2) according to the modified device design, carrying out the parametric test on the prototype devices (2) to obtain a new set of parametric test result values and applying the method to the new set of parametric test result values, anda step of adjusting a target yield.

10. Method according to any one of the preceding claims,wherein obtaining (13,15,16) the set of parametric test result values comprisesobtaining (13) a raw data set of parametric test result values, applying at least one anomaly detection technique to the raw data set, andremoving parametric test result values determined to be anomalies from the raw data set.M / ADVA-017-PC 2111. Method according to any one of the preceding claims, further comprising deploying (27) the obtained ones of the upper specification limit value (USL) and the lower specification limit value (LSL) to automated test equipment for use in a production environment for producing the devices (2).

12. Method according to claim 11, further comprisingcarrying out parametric testing and an acceptance test using the deployed specification limit values (USL, LSL) on devices (2) produced in the production environment.

13. System for generating at least one of an upper and a lower specification limit value for use as acceptance criterion in a device acceptance test, comprising:an interface (14) for obtaining parametric test result values; an interface (14) for returning the generated specification limit values; anda data processing system (10), configured to carry out a method according to any one of the preceding claims.

14. Computer program comprising instructions which, when the program is executed by a computer (10) comprising an interface (14) for obtaining parametric test result values and an interface (14) for returning specification limit values, cause the computer to carry out a method according to any one of claims 1-12.

15. Computer-readable medium having stored thereon a computer program according to claim 14.

Images