CN102253609B - Method for processing alignment mark measuring signals - Google Patents

Abstract

The invention provides a method for processing alignment mark measuring signals. The method comprises the following steps of: after sampling an alignment signal, fitting the signal by a fitting model to solve the best alignment position; performing model fitting by using a result acquired by light intensity sampling and position sampling; fitting the period of the signal while fitting model parameters so as to determine the phase of the signal; and finally solving and determining the best alignment position.

Description

Method for processing alignment mark measuring signals

Technical field

The present invention is relevant with the lithographic equipment that integrated circuit or other microdevice are manufactured field, particularly a kind of method for processing alignment mark measuring signals.

Background technology

In semiconductor work C ic manufacturing process, complete chip need to just can complete through photolithographic exposure repeatedly conventionally.Except photoetching for the first time, before exposure, the photoetching of all the other levels figure staying that all figure of this level and former level will be exposed is accurately located, to guarantee the correct relative position between two layer patterns, i.e. alignment.Overlay error only allows within 1/3 scope of photolithography resolution conventionally.Affect the many factors of alignment precision, comprise the positioning precision of work stage, the measuring error of position measuring system, alignment error, the alignment error of machine etc. of mask and wafer, wherein the alignment precision between silicon chip different process layer is also one of important influence factor.Due to after silicon chip completes one deck graph exposure on lithographic equipment, need to descend the processing of the follow-up semiconductor technologies such as sheet cures, development, and then again go up slice to lithographic equipment, carry out the photoetching of next layer pattern.Before carrying out this layer pattern photoetching, first-selection need to be passed through alignment system, sets up the position coordinates relation between this layer of mark and last layer mark, alignment relation accurately between guarantee two figures.In actual production, for fear of the propagation of error that mark is aimed between layers, often adopt zero layer of mark as reference mark.Each layer of mark aimed at zero layer of mark, sets up the position coordinates relation between each layer of mark and zero layer of mark.In the present invention, the aligning of mark is the measurement of mark, and registration signal is the measuring-signal that measurement markers obtains.

Patent CN03164859.2, CN200710045495X, CN2007100454964 and US6297876B1 etc. have introduced silicon chip (from the axle) alignment system of a class based on optical grating diffraction.This alignment system adopts the alignment mark (as 16 microns and 17.6 microns) that comprises two different cycles sub-gratings, by survey two sub-gratings ± 1 grade of interference of light picture sees through the light intensity signal with reference to grating, through the matching of signal, comprehensively determine the coarse alignment position of mark; Meanwhile, utilize the senior interference of light imaging (as ± 5 grades of light) of meticulousr grating ± 1 grade interference of light picture or 16 microns of periodic optical gratings, through the matching of signal, on coarse alignment (measurement) basis, determine fine alignment (measurement).In such patent, the matching of registration signal is the phase place of determining signal, conventionally adopts following model of fit:

In formula, I is light intensity, and x is position, coefficient a ₀, a ₁..., a _i, b ₀, b ₁..., b _jand phase place

for model undetermined parameter, the cycle that P is signal, need predetermined constant.Matching be by positions different in registration signal and luminosity sampling pair, and signal period P, determines the undetermined parameter of model by least square fitting.In patent CN200510030807.0, CN200810035115.9, CN200910055927.4, CN200510030577.8, signal period P is pre-set machine constant.This machine constant can utilize design load directly to introduce, or demarcates acquisition by surveying school.Alignment mark on datum plate is little with environmental change, and the cycle of the registration signal of generation is also substantially constant, and it is feasible adopting fixing signal period P to carry out models fitting.But, for the alignment mark on silicon chip, be subject to processing technique and environmental change impact large (as the thermal expansion of silicon chip etc.).Different process layer mark on mark on different silicon chips, same silicon chip, and the cycle of the registration signal that on same silicon chip, zones of different mark produces be all different.Now, adopt same fixed signal cycle P to carry out models fitting as continued, will cause fitness bias.As shown in Figure 1, now the actual cycle of registration signal was not mated with the matching cycle, and actual signal and matched curve deviation are very large.In addition, in the time there is deviation in signal actual cycle and matching cycle, peak point (obtaining from matched curve) will with sampling point distributions Existence dependency, utilize the different section matchings of registration signal, it is different obtaining a series of peak points.And registration signal matching section is closely related with the prealignment position that the last process of aiming at provides mark, thereby make alignment result have very large uncertainty.

In the present invention, signal period P does not introduce as fixed value, but participates in matching as a parameter for the treatment of matching.Because the now matching cycle solves acquisition by matching, rather than directly given as machine constant, to not have matching cycle in technology formerly and the inconsistent problem of actual cycle of signal, the matched curve of acquisition is also by more identical with the actual conditions of signal, phase place solving precision will obtain raising, and then reduce alignment error.

Summary of the invention

The object of the present invention is to provide a kind of new method for processing alignment mark measuring signals, the signal (comprising luminosity sampling and position sampling) that this signal processing system obtains measurement markers carries out models fitting, the process of matching is carried out matching to the cycle of signal simultaneously, thereby determines the phase place of signal.

Method for processing alignment mark measuring signals provided by the invention utilizes following model of fit to carry out signal fitting and solves best aligned position after utilizing alignment system to sample to registration signal:

In formula, I is light intensity, and x is position, a ₀, a ₁..., a _i, b ₀, b ₁..., b _jfor model coefficient, i, j is non-negative integer,

for phase place, the cycle that P is signal;

Utilize light intensity and position sampling, carry out determining model coefficient a after numerical solution ₀, a ₁..., a _iwith b ₀, b ₁..., b _j, signal period P and phase place

value.

Wherein, described method of value solving adopts successive approximation approach.

Wherein, described successive approximation approach adopts Gauss-Newton method or Marquardt method.

Wherein, given first model coefficient a in solution procedure ₀, a ₁..., a _iwith b ₀, b ₁..., b _j, signal period P, phase place

initial value a ₀ ⁽⁰⁾, a ₁ ⁽⁰⁾..., a _i ⁽⁰⁾, b ₀ ⁽⁰⁾, b ₁ ⁽⁰⁾..., b _j ⁽⁰⁾, P ⁽⁰⁾,

and near this initial value, model of fit is carried out to taylor expansion, and omit second order and above higher order term, model of fit is converted into linear equation and is:

In formula,

$\frac{\∂I}{\∂a_i}\mathrm{\Δ}{a}_i=x^i\mathrm{\Δ}{a}_i,$ i＝0，1，2，…，

j＝0，1，2，…，

Δ a _i, Δ b _j, Δ P and

poor between true value and initial value, that is: Δ a _i=a _i-a _i ⁽⁰⁾, Δ b _j=b _j-b _j ⁽⁰⁾, Δ P=P-P ⁽⁰⁾,

now, the parametric solution of model of fit turns and just turns to Δ a _i, Δ b _j, Δ P and

linear equation solve.

Wherein, a _i, b _j,

initial value determined by experience or at random determine, the initial value of the cycle P of signal is according to the design parameter determination of alignment system used.

Wherein, utilize least square method to iterate and solve Δ a _i, Δ b _j, Δ P and

until calculate | Δ a _i|, | Δ b _j|, | Δ P| and

be less than threshold value separately.

Wherein, described threshold value determined by experience, or according to a after first round iteration is upgraded _i ⁽⁰⁾, b _i ⁽⁰⁾, P ⁽⁰⁾,

value determine.

In the disposal route that the present invention adopts, the cycle P of signal is also as treating fitting parameter, with a ₁, a ₂..., a _i, b ₁, b ₂..., b _kwith

carry out together matching, mate the models fitting under the cycle thereby obtain most with registration signal, and then can effectively improve aligning (measurement) precision of silicon chip mark.

Accompanying drawing explanation

Figure 1 shows that the matching cycle affects schematic diagram to signal accuracy;

Figure 2 shows that the alignment system that the present invention is used.

Embodiment

Below, describe in detail according to a preferred embodiment of the invention by reference to the accompanying drawings.For convenience of description and highlight the present invention, in accompanying drawing, omit existing associated components in prior art, and by the description of omitting these well-known components.

Figure 2 shows that signal processing method according to the present invention alignment system used.This alignment system comprises: electric light source and lighting module 1; Image-forming module 2; With reference to grating 3; Signals collecting and processing module 4; Alignment mark 5, comprises datum plate mark and is positioned at the silicon chip mark on silicon chip 6; Work stage 7; Sports platform 8; Station acquisition and motion-control module 9; Alignment function and administration module 10.Electric light source and lighting module 1 provides illumination beam to alignment mark, forms the diffraction light that carries label information, and diffraction light is imaged onto with reference on grating 3 surfaces by image-forming module 2.Station acquisition and motion-control module 9 gather the positional information of the work stage 7 of carrying silicon chip, and synchronize and negotiate with alignment function and administration module 10, planning movement locus, the motion of controlled motion platform 8, meanwhile, the positional information for aiming at is transferred to after treatment to alignment function and administration module 10.The motion of sports platform 8 makes alignment mark imaging scan reference grating 3 and produces light intensity signal.Signals collecting and processing module 4 gather light intensity signal, and are transferred to after treatment alignment function and administration module 10.Alignment function is comprehensive from the light intensity signal of multiple mark sub-gratings and the positional information in order to aim at administration module 10, through signal fitting, calculates the aligned position of asking for mark.Concrete alignment procedures can, with reference to formerly patent CN200710045495X, CN2007100454964 etc., be introduced as known technology here.

Fitting within alignment function and administration module 10 of registration signal carried out, and adopts as drag:

In formula, I is light intensity, and x is position, a ₀, a ₁..., a _i, b ₀, b ₁..., b _jfor model coefficient, i, j is non-negative integer,

for phase place, the cycle that P is signal.

The object of registration signal matching is exactly to obtain light intensity and position sampling according to alignment scanning process, determines the phase place of signal

and and then determine a series of peak points of this registration signal.Formerly in patent CN200510030807.0, CN200810035115.9, CN200510030577.8, only model coefficient a ₀, a ₁..., a _i, b ₀, b ₁..., b _jand phase place as treating that fitting parameter solves, and signal period P directly introduces as pre-set fixed value.In the time there is deviation in the predefined fixed value of P and signal actual cycle, the cycle of the matched curve obtaining by matching and the actual cycle of registration signal inconsistent (seeing Fig. 1), the phase place obtaining

error also will be become greatly, and then affect alignment precision.In the present invention, signal period P is also as treating fitting parameter, with model coefficient a ₀, a ₁..., a _i, b ₀, b ₁..., b _jand phase place

solve together by matching, can guarantee that the cycle of matched curve and the cycle of actual signal that obtain are consistent, also improved phase place simultaneously

solving precision, and then improved alignment precision.

Because formula (2) is nonlinear function, cannot direct solution.In numerical evaluation, the method processing that employing successively approaches conventionally, as Gauss's one Newton method, Marquardt method etc.The present invention has only provided the embodiment that Gauss-Newton method is processed, and those skilled in the art can adopt other method for solving to process with condition as required.

Solution procedure given first model coefficient a ₀, a ₁..., a _iwith b ₀, b ₁..., b _j, signal period P, phase place

initial value a ₀ ⁽⁰⁾, a ₁ ⁽⁰⁾..., a _i ⁽⁰⁾, b ₀ ⁽⁰⁾, b ₁ ⁽⁰⁾..., b _j ⁽⁰⁾, P ⁽⁰⁾,

and near this initial value, formula (2) is carried out to taylor expansion, and omit second order and above higher order term, formula (2) is converted into linear equation:

$\frac{\∂I}{\∂a_i}\mathrm{\Δ}{a}_i=x^i\mathrm{\Δ}{a}_i,$ i＝0，1，2，…

j＝0，1，2，…

Here Δ a, _i, Δ b _j, Δ P and

poor between true value and initial value, that is: Δ a _i=a _i-a _i ⁽⁰⁾, Δ b _j=b _j-b _j ⁽⁰⁾, Δ P=P-P ⁽⁰⁾,

the parametric solution of formula (2) is also just converted into Δ a _i, Δ b _j, Δ P and solve.

When alignment scanning, can obtain a series of sampled point in registration signal, measurement sampled point as shown in Figure 1, these measurement point composition light intensity-position samplings are to { I _n, x _n, n=1,2,3 ...By least square method, easily solve Δ a _i, Δ b _j, Δ P and

value.When | Δ a _i|, | Δ b _j|, | Δ P| and

while being greater than threshold value separately, renewal initial value is: a _i ⁽⁰⁾=Δ a _i+ a _i, b _j ⁽⁰⁾=Δ b _j+ b _j, P ⁽⁰⁾=Δ P+P,

then re-use least square and solve Δ α _i, Δ b _j, Δ P and

so iterate, until | Δ a _i|, | Δ b _j|, | Δ P| and

be less than threshold value separately.A now _i ⁽⁰⁾, b _j ⁽⁰⁾, P ⁽⁰⁾,

be final solution.The phase place of matching acquisition registration signal is so:

the cycle of registration signal is: P=P ⁽⁰⁾.

Peak value (crest) position is:

k＝…，-1，0，1，…

In this embodiment, a _i, b _j,

initial parameter can be determined by experience, also can determine at random, for example, set a _i ⁽⁰⁾=0, b _i ⁽⁰⁾=0, the initial value of cycle P is according to design parameter determination.For example, for the grating marker of 8.0 microns of live widths, its ± theoretical value of 1 grade of optical registration signal is 8.0 microns, therefore set P ⁽⁰⁾=8 microns.

Δ a _i, Δ b _j, Δ P and

threshold value can be determined by experience, also can according to through one take turns iteration upgrade after, a now _i ⁽⁰⁾, b _j ⁽⁰⁾, P ⁽⁰⁾,

value determine.For example, set Δ a _ithreshold value is a _i ⁽⁰⁾/ 10, other threshold value is similar.

Described in this instructions is several preferred embodiment of the present invention, and above embodiment is only in order to illustrate technical scheme of the present invention but not limitation of the present invention, and for example, the present invention also can apply on double-workpiece-table lithographic equipment.All those skilled in the art, all should be within the scope of the present invention under this invention's idea by the available technical scheme of logical analysis, reasoning, or a limited experiment.

Claims (7)

1. a method for processing alignment mark measuring signals, is characterized in that after utilizing alignment system to sample to registration signal, utilizing following model of fit to carry out signal fitting solves best aligned position:

In formula, I is light intensity, and x is position, a ₀, a ₁..., a _i, b ₀, b ₁..., b _jfor model coefficient, i, j is non-negative integer,

for phase place, the initial value that P is the signal period;

Utilize light intensity and position sampling, carry out determining model coefficient a after numerical solution ₀, a ₁..., a _iwith b ₀, b ₁..., b _j, signal period P and phase place

value.

2. method for processing alignment mark measuring signals according to claim 1, is characterized in that: described method of value solving adopts successive approximation approach.

3. method for processing alignment mark measuring signals according to claim 2, is characterized in that: described successive approximation approach adopts Gauss-Newton method or Marquardt method.

4. method for processing alignment mark measuring signals according to claim 3, is characterized in that:

Given first model coefficient a in solution procedure ₀, a ₁..., a _iwith b ₀, b ₁..., b _j, signal period P, phase place

initial value a ₀ ⁽⁰⁾, a ₁ ⁽⁰⁾..., a _i ⁽⁰⁾, b ₀ ⁽⁰), b ₁ ⁽⁰⁾..., b _j ⁽⁰⁾, P ⁽⁰⁾,

and near this initial value, model of fit is carried out to taylor expansion, and omit higher order term more than second order, be converted into Δ a _i, Δ b _j, Δ P and

linear equation:

In formula,

Δ a _i, Δ b _j, Δ P and poor between true value and initial value, that is: Δ a _i=a _i-a _i ⁽⁰⁾, Δ b _j=b _j-b _j ⁽⁰⁾, Δ P=P-P ⁽⁰⁾,

5. method for processing alignment mark measuring signals according to claim 4, is characterized in that: a _i, b _j,

initial value determined by experience or at random determine, the initial value of the cycle P of signal is according to the design parameter determination of alignment system used.

6. method for processing alignment mark measuring signals according to claim 5, is characterized in that: utilize least square method to iterate and solve Δ a _i, Δ b _j, Δ P and

until calculate | Δ a _i|, | Δ b _j|, | Δ P| and

be less than threshold value separately.

7. method for processing alignment mark measuring signals according to claim 6, is characterized in that: described threshold value determined by experience, or according to a after first round iteration is upgraded _i ⁽⁰⁾, b _j ⁽⁰⁾, P ⁽⁰⁾,

value determine.

Images