1. Introduction

Simple and effective methods for inspection of objects with characteristic dimensions of several dozen nanometers are required by the modern semiconductor industry [1]. Conventional measuring tools such as the scanning tunneling microscope (STM) and atomic force microscope (AFM) provide near-atomic level imaging resolution but cannot be used in a mass production environment due to low throughput and the destructive nature of inspection. This is particularly true during quality control steps, where fast inspection of nanochips is required. Critical dimension (CD) scanning electron microscope-based CD-SEM technology is limited to CD measurements only, and cannot be applied to side wall angle (SWA), undercut, or compositional simultaneous measurements for 3D FinFET topologies [2].

In order to solve this problem, several optical methods with comparative simplicity and high performance have been developed [3, 4]. The recently proposed through-focus scanning optical microscopy (TSOM) method [5–8] is promising as an inspection tool for semiconductor manufacturing processes. This method offers optical inspection of non-periodic and isolated nanoscale objects with sub-nanometer accuracy, restricted by the light detector noise threshold [9]. However, mechanical positioning instabilities arise because TSOM requires the mechanical scanning of a sample. These instabilities result in additional noise, and hence put restrictions on measurement accuracy. As positioning instabilities tend to increase with an increase in mechanical scanning speed, throughput is also restricted.

Thus, it is desirable to develop all optical techniques that possess all the advantages of TSOM and lack mechanical scanning.

In our work, we demonstrate an all optical method of inspection using a specially designed low cost light source tunable in the light emitting diode bandwidth with a spectral width of about 30 nm and an optical system with chromatic aberration. Chromatic aberration utilization has been already demonstrated for other microscopy applications [10], and in our work such a combination of tunable light source and optical system chromatic aberration provides defocusing without mechanical motion in the required range. This approach thus allows one to capture motion-free, through-focus diffraction patterns. This method has about the same accuracy as TSOM methods based on mechanical scanning of a sample but has the potential advantages of stability and precision measurements at high throughput for industry applications.

2. Methods

2.1 System introduction

The optical scheme of our method is presented in Fig. 1. This setup utilizes a bright field microscopy scheme with Kohler illumination [11]. The fiber output of our custom designed tunable light source is imaged via a relay system composed of two lenses (L1, L2) to the back focal plane of a single component objective lens (glass: ECO-550, f = 4 mm, NA = 0.6) that possesses chromatic aberration. Images are captured by the CCD matrix (4.65 × 4.65 μm² pixel size, 12 bit) with 62x magnification from an Ophir-Spiricon Beam Gage laser beam analyzer. The only difference from TSOM scheme is utilization of the tunable light source with moderate spectral stability which allows defocusing without mechanical motion of the sample due to objective lens chromatic aberration. Effect of aberrations, misalignment and light source power stability is comparable in both approaches.

 

Fig. 1 Experimental setup (right). Tunable light source setup (left).

Download Full Size | PDF

A light source was designed to provide a 30 nm range of wavelength tuning with central wavelength 455 nm. In combination with a single component objective lens, this setup resulted in ± 15 µm of focus offset. In our work we use light source tunable by mechanical means for approach demonstration, however different types of tunable light source including laser diodes with temperature depended wavelength. The light source design is based on a diffraction grating monochromator scheme (Fig. 1, left) and consists of a multimode fiber coupled light source (Fiber1: d = 600 µm, NA = 0.4, I = 8 mW, λ = 455 nm, Δλ = 30 nm) working as an entrance slit, and a reflective ruled diffraction grating with 1200 l/mm and a 50x50 mm aperture. A multimode fiber (Fiber2: d = 400 µm, NA = 0.6) is used as the output slit of the monochromator. Both fibers are collimated via lenses F1, F2 (F1 = 125 mm, F2 = 88 mm, D1 = D2 = 50 mm). The finished scheme provides tunable illumination in the range 440-470 nm with a spectral width Δλ~4 nm as defined by the light source size (Fiber1). The light power at the fiber output (Fiber2) is about 100 µW. For wavelength scanning defocusing measurements, a diffraction grating is placed on a precise Newport Conex motorized rotation stage with closed loop control, developed using LabVIEW software.

2.2 Experiment

With this approach, we performed measurements with a calibrated sample fabricated by NTT Advanced Technology Corporation. The sample has 12 lines with the outside area etched in a monocrystalline silicon substrate with a 40 µm gap between adjacent lines. Geometrical dimensions are: height 50 nm, length 100 μm, and width (the parameter under study or CD) 40−150 nm (deviation less than 2 nm) with 10 nm steps. SEM images of the samples with 3 different CD values are presented in Fig. 2.

 

Fig. 2 SEM images of test objects.

Download Full Size | PDF

Experimental measurements were performed as follows: test objects were positioned to best approximate the focus for the central wavelength of 455 nm prior to measurements using the position system. The rotation angle of the motorized positioning stage with the diffraction grating was set to 10°, which corresponds to the middle of the working range of wavelength tuning. The rotation angle of the grating was then scanned from 9° to 11° with the 0.01° steps. Each grating rotation angle corresponds to a different average illumination wavelength. The objective lens chromatic aberration results in the appropriate focus offset. This dependence is captured by measuring the change in focus position required to return back to the best focus position. Corresponding calibration curves for the focus offset without mechanical motion and the illumination intensity dependence of the rotation angle are shown in Fig. 3 and were used for data recalibration. For each diffraction grating angle position corresponding to the appropriate sample focus offset, a sampled diffraction pattern with 232x232 pixel² size was captured. As the position of each sample in the field of view is the same during measurements, we exclude the possible influence of changing illumination conditions across the field of view.

 

Fig. 3 Experimental setup calibration. (a) Normalized dependence of illumination intensity on grating angle value. (b) Sample focus offset dependence on grating angle value. The straight line is a linear fit to the measured data.

Download Full Size | PDF

2.3 Data analysis

The next step includes the averaging of diffraction patterns along the y-axis parallel to the silicon line direction (Fig. 2), with subsequent construction of a through-focus diffraction pattern. Typical normalized through-focus diffraction patterns, obtained for the two test objects 40 nm and 50 nm in width object are shown in Figs. 4(a) and 4(b). Figure 4(c) demonstrates the differential through-focus diffraction pattern image for 40 nm and 50 nm CD values with intensity well exceeding the noise threshold. Normalized through-focus diffraction patterns for a 40 nm CD value using direct mechanical scanning by a piezo-mechanical system as normally used in TSOM technology are presented in Fig. 4(d). One can see the qualitative correspondence between diffraction patterns measured using mechanical and mechanical-free approaches. This fact allows us to state the equivalence of the original TSOM and modified motion-free approaches, and apply all results related to experimental and simulation data coincidence demonstrated in [9] to the proposed technology.

 

Fig. 4 Test object measurements. (a) Normalized through-focus diffraction pattern obtained for an object with a 40 nm CD. (b) Normalized through-focus diffraction pattern for an object with a 50 nm CD. (c) Differential through-focus diffraction pattern for test objects with a 40 nm and 50 nm CD. (d) Normalized through-focus diffraction pattern obtained with a mechanical scanning approach.

Download Full Size | PDF

In order to compare objects with different CD values, we define a metric for each object similar to [9]. First, the subtraction of the normalized through-focus diffraction pattern of the reference object from that of the test object was performed. The resulting pattern is referred to as differential through-focus diffraction pattern (DTDP). The metric value is further defined as the sum of absolute values of all pixels within the DTDP, normalized by the total number of pixels. In our study, the size of through-focus diffraction patterns was the same and diffraction pattern for the object with a 40 nm CD value was used as a reference

Figure 5 demonstrates the measured dependence of the metric on the CD value of the test objects. The curve is monotonic and can be used for metrological applications with silicon lines based on either a measured or simulated library of diffraction patterns corresponding to similar measurement conditions.

 

Fig. 5 Change in metric with CD value.

Download Full Size | PDF

Using the metric curve, we can estimate measurement accuracy in the presence of the measurement errors such as camera noise, vibrations, etc.

A series of measurements of the metric for each test object was carried out and the standard deviation σ was estimated. The spread of these values, calculated as ± 2σ, determined the total measurement error. The dashed pairs of lines in Fig. 5 show the ± 2σ deviation range due to measurement errors. The obtained data allow an evaluation of the measurement accuracy limit as ~1 nm for Si lines with a width of 40−60 nm. The same evaluation was obtained in [9] using a mechanical scanning approach.

The test of this metrology tool in repeated measurements based on measured libraries for 12 silicon samples demonstrates the confident recognition of objects with CDs from 40−90 nm and 130−150 nm. Erroneous recognition was found for objects with CDs in the range of 100−120 nm due to the flatness of the metric curve in this range. In most cases 10% CD value accuracy is of practical interest, so for every CD inspection range, similar to original TSOM method, the illumination and measurement conditions should be optimized to maximize the sensitivity.

3. Conclusions

A method for all optical inspection of nanoscale objects without mechanical scanning of the sample position is demonstrated. Such a method is promising for stable and precise nanoscale inspection. This method lacks the mechanical instabilities inherent in sample positioning caused by a mechanical positioning stage, and allows an increase in throughput due to the ability to quickly and precisely tune the wavelength. An expensive tunable laser source is not required for this measurement system. We demonstrated that available light sources with a narrow range of tuning wavelength ~10−30 nm, including laser diodes, can be used for these measurements. The proposed method is tested using objects with different CD values within the 40−150 nm range. Results are compared to those obtained with a mechanical inspection system based on the direct defocusing of the object with a piezo-nanopositioning stage. A similar CD measurement accuracy limit was achieved for both systems, corresponding to about 1 nm.