DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0108] One aspect of the invention is removal of a microscopic target structure which is part of a multilayer, multimaterial device, wherein laser energy is incident on several materials having dissimilar optical and thermal properties. One application is memory repair. A new fabrication process (Damascene) includes a copper target structure, multiple dielectric layers in the form of a “stack,” and functional circuitry disposed at the dielectric layers. The target structure and layers are typically formed on a silicon substrate. This is illustrated in FIGS. 1 b and 1 c and corresponds to a device processed with an embodiment of the present invention. This will be referred to as a “multilevel” process.

[0109] With the use of more complex structures at finer scale (e.g., at or below a wavelength of visible light), considerations for reliable operation of laser processing system increase to meet the standards for high yield in the semiconductor industry.

[0110] Aspects of the invention include methods and subsystems for operation of the laser processing system. At the microscopic scale, the laser beam waist diverges rapidly due to the small spot size and depth of focus. The materials within the 3D beam location may include functional circuitry. In an automatic system, robust measurement of target locations is used in conjunction with database information to position a laser beam in three dimensions at high speed. The interaction of a laser beam within the multilevel device influences yield. Modeling of thermal interaction is useful of understanding and predicting performance in the thermal processing regime. However, at the microscopic scale, a more detailed understanding of interaction based on physical optics is also beneficial.

[0111] In the following sections, detailed aspects of spatial and temporal pulse shaping, three-dimensional measurement and prediction, device modeling and process design are disclosed with emphasis on solving the problem of cleanly removing links on a multilevel device, wherein damage is avoided to inner layers and functional circuitry between a link and the substrate. However, various methods, subsystems, and experimental results may also be applied for link processing of conventional single inner layer devices, and generally for processing microstructures surrounded by materials having dissimilar thermal or optical properties.

[0112] Processing Links on a Multilevel Device

[0113] A pulsed laser beam, the beam having pre-determined characteristics for processing of microscopic structures, is used to cleanly remove at least a portion of a target structure. An application of the method and system of the present invention is severing of highly reflective copper links which are part of a high speed semiconductor memory device. The method and system of the present invention is particularly advantageous for processing of targets having a sub-micron dimension, including targets with a dimension below the wavelength of the laser beam. The target is separated from a semiconductor substrate by a multi-layer stack, which may have several dielectric layers. Furthermore, both the temporal and spatial characteristics of the pulse may be selected or controlled based on the thermal and optical properties of the microscopic target, underlying layer materials, and the three-dimensional layout of the device structure, including the spacing of target structures and functional inner conductor layers.

[0114] FIGS. 1 a - 1 c generally show an embodiment of the present invention A laser pulse 3 irradiates a rectangular target structure or microstructure 10 , side views of which are shown in FIGS. 1 b and 1 c , with a focused beam. In a preferred embodiment, an output from short pulse amplified laser system 1 is generated to produce the pulse 3 which has a rise time 4 fast enough to efficiently couple energy into a highly reflective target structure. The duration 5 is sufficient to process the target structure wherein at least a portion of the structure is cleanly removed without leaving residue, slag, or other debris. The fall time 6 is preferably fast enough to avoid creating undesirable damage to the layers or substrate.

[0115] The temporal pulse shape is selected, in part, based on physical properties of the target microstructure 10 , for instance, thickness, optical absorption, thermal conductivity, or a combination thereof. In an advantageous embodiment of the invention, the processing will occur with a single pulse having a fast edge leading relative to a selected pulse duration of several nanoseconds. In an alternative embodiment, the laser output may be a series of narrow q-switched or rectangular pulses, with very fast rise time, for example 800 ps pulses representative of the output of commercially available q-switch micro-lasers. The pulses may be delayed with respect to each other so as to provide a burst of pulses to irradiate the target structure. The laser output may be generated with a combination of a high bandwidth seed laser diode and fiber optic amplifier with Raman shifting, or with a waveguide amplifier system. Alternatively, a desirable pulse characteristic may be provided with various modified q-switched systems or with the use of high speed electro-optic modulators. Other pulse shapes may be selected for the material processing requirements. For instance, a sequence of closely spaced pulses having duration from a few picoseconds to several nanoseconds is taught in Reference 5.

[0116] In one embodiment, a high bandwidth MOPA configuration is used to amplify the laser output of a high speed semiconductor diode. Generation of various pulse shapes and duration with direct modulation of the diode is considered advantageous, provided any affect associated with variable amplitude drive waveforms does not affect overall performance. Further details of various aspects of pulse generation and amplification can be found in references 5 and 6 (e.g., in '471—Reference 5—FIGS. 5 and columns 14-16).

[0117] As indicated above, embodiments of the laser system may include fiber optic amplifiers which amplify the preferred square pulse shape generated by a seed laser. The seed laser may be a high speed semiconductor diode or the shaped output of a modified q-switched system. The amplified output may be matched in wavelength to the input or Raman-shifted as taught in References 4 and 6 (e.g., in Reference 6, FIGS. 12-13 and column 14, line 57—column 19, line 3). Wavelength shifting of a short pulse q-switched laser output is generally taught in '759 Reference 4.

[0118] In an alternative arrangement the seed laser is a semiconductor diode and the optical amplifier is a waveguide amplifier. Advantages of an embodiment with a waveguide amplifier when compared to a fiber system include avoidance of Raman shifting, lower pulse distortion at the speed of operation, and, with proper design, minimal thermal lensing. A precision anamorphic optic system is used to optimize coupling between the seed and amplifier. Basic description of waveguide amplitude and lasers can be found in product literature provided by Maxios, Inc. and in the article “CW and passively Q-switched Cladding Pumped Planar Waveguide Lasers,” Beach et. al. Yet another amplifier system including a 28 DB planar waveguide amplifier for use at 1.064 μm wavelengths was developed by University of Southhampton and described in “A Diode Pumped, High Gain, PlanarWaveguide, Nd:Y3Al5O12 Amplifier.”

[0119] In an alternative arrangement, for generation of a fast rising pulse or other desirable shape, a plurality of q-switched micro-lasers can be used. The modules produce a q-switched waveform with pulse durations of about 1 nanosecond or less, for example 800 ps to 2 ns for commercially available units. An example of a commercially available laser is the AOT-YVO-1Q available from Advanced Optical Technology (AOTLasers.com). These recently developed short pulse, active q-switch lasers can be triggered with a TTL pulse at a variable repetition rate while maintaining specified sub-nanosecond timing jitter. In general, the pulse shape incident on the target microstructure will vary significantly at repetition rates approaching the maximum rate. Reference 9 teaches methods of maintaining a constant pulse shape despite variations in the temporal spacing of pulses incident on a target (e.g., see the figures and associated specification). AOT offers a pulsewidth of 2 nanoseconds available at a repetition rate of 20 KHz. Frequency doubled versions are also available (532 nm). IMRA America reports 800 ps pulses with the PicoLite system, and high peak power was obtained with fiber amplification at repetition rates up to 10 KHz. Shorter pulsewidths, for instance about 1 ns or less, are available at slower repetition rates.

[0120] As known in the art and illustrated in Reference 5 (e.g., FIGS. 1 c , 2 ), the q-switched waveforms may approximate (at least to 1st order) a symmetric Gaussian shape, or a fast rising pulse with an exponential tail, depending on the stored energy. With reference to FIGS. 15 a - 15 c , a series of devices, with appropriate delays introduced by a plurality of triggering signals, or delays of a trigger signal with a delay line, is used to generate a series of spaced apart pulses. The optical outputs are preferably combined with appropriate bulk optics (polarization sensitive), fiber optics, or waveguides to form a single output beam. The resultant addition of the q-switched waveforms produces a fast rise time characteristic and relatively short duration. An optical amplifier 122 may be used to increase the output power as needed.

[0121] FIG. 15 a shows a schematic of one basic embodiment with bulk optics, where a beam combiner 123 is used to deliver the output of two lasers 120 , 121 to an amplifier 122 . A delay circuit 126 , which may be programmable, controls triggering. Polarization optics 127 , 128 are used to provide the proper input to the beam combiner. In one arrangement the pulses are spaced apart and appear as a high frequency burst 124 . In a second arrangement triggering of the second pulse occurs at a slightly delayed (but controlled) position which produces a characteristic approximating a square pulse shape 125 . In the latter arrangement the controlled delay is about 50% of the FWHM. Those skilled in the art will recognize that alternative arrangements may be used with multiple amplifiers, combiners, with bulk, fiber, or integrated optic arrangements.

[0122] Generation of multiple pulse waveforms may also include some form of active q-switching of two separate microlasers or detecting a first pulse from a passively q-switched laser and subsequently triggering an actively q-switched laser or MOPA relative to the first pulse.

[0123] FIG. 15 - b is a basic schematic showing the use of a single laser 140 wherein the laser output is divided by beam splitter 142 , whereby a portion of the beam propagates along a path 141 , followed by combining with combiner 143 , after polarization adjustment with rotator 146 which may be a half-wave plate. An optional optical amplifier 145 may then be used to produce higher output power.

[0124] In an arrangement using a single laser and an optical delay line, the optical system will preferably be stable and easy to align. FIG. 15 c shows an exemplary embodiment wherein the use of opposing corner cube retroreflectors 130 makes the setup insensitive to tilt of the folding elements. The angular alignment of the delayed beam paths 131 , 132 is very stable even in a high vibration environment. One of the corner cubes in each pair of retroreflectors 130 is initially adjusted in the X/Y translation and Z rotation to get the transverse position of the delayed beam path centered. Each of the λ/2 retarders 133 in the main beam path is adjusted so that vertical or horizontally polarized light will have its polarization rotated by 45 degrees. The λ/2 retarder 133 in the second delay loop is adjusted so that vertical or horizontal polarized light will have the polarization rotated by 90 degrees causing the delayed pulse in the second loop to circulate twice before exiting. The peak-to-peak spacing of the output waveform 135 (e.g., 4 combined pulses) is controlled by the length of the delay loops. If non-equal amplitudes for the delayed pulses are desired, the λ/2 retarders 133 in the main beam can be set for a polarization of other than 45 degrees. Likewise, the pulse shape can be varied at the time a system is setup or possibly in operation by manually or automatically controlling the spacing. Those skilled in the art of laser pulse generation and shaping will appreciate the advantages of the compact and modular arrangement for short pulse for typical delays ranging from a few nanoseconds to tens of nanoseconds. For instance, U.S. Pat. No. 5,293,389 to Hitachi describes a polarization-based fiber delay line for generating laser pulses of decreasing amplitude for generating longer pulses, for instance 100 ns or longer.

[0125] Another means of producing a shaped pulse is to use the modulator approach to chop the leading edge or tail of the pulse but with a two-stage or shaped modulation voltage pulse. For example: with a 10 ns q-switched pulse, the modulator could have 100% transmission for the first 1-5 ns followed by 25% transmission for the remainder of the pulse. Early pioneering work by Koechner (U.S. Pat. No. 3,747,019) and Smart (U.S. Pat. No. 4,483,005) demonstrate exemplary amplitude and pulse shape control methods using electro-optic modulators.

[0126] The multiple pulses shown in FIGS. 15 a - 15 c may or may not have the same wavelength, and the temporal shape of a pulse may be varied depending upon specific requirements. For example, in certain embodiments an output may be a q-switched pulse of short duration and high peak power combined with a lower power square pulse shape.

[0127] Referring to FIGS. 1 a and 1 b , during system operation for memory repair, position information, obtained with a precision measurement system, is used to relatively position the focused beam waist of the pulsed laser at a location in space 7 , 8 , 9 to substantially coincide with the target 10 three-dimensional coordinates (Xlink,Ylink,Zlink). A trigger pulse 2 , generated at a time where the laser beam waist and target position substantially coincide, operates in conjunction with the laser and associated control circuitry in laser subsystem 1 to produce an output pulse.

[0128] References 2 and 7 describe details of a method and system for precision positioning, including three-dimensional beam waist positioning. Reference 7 describes a preferred embodiment for producing an approximate diffraction limited spot size with a range of spot size adjustment (e.g., FIGS. 7-9of WO0187534 ('534) and the associated specification), and a preferred method and system for three-dimensional positioning of the beam waist. Three-dimensional (height) information is obtained, for instance with focus detection, and used to estimate a surface and generate a trajectory (e.g., FIGS. 2-5 of '534 and the associated specification). The laser is pulsed at a location substantially corresponding to the three-dimensional position of the link (Xlink, Ylink, Zlink) (e.g., FIGS. 10 a - b of '534 and the associated specification).

[0129] In practice, the three-dimensional measurement and positioning are used to compensate for topographical variations over a wafer surface, or other position variations introduced in a system (mis-alignment). These variations are generally system or application dependent and may exceed several microns, which in turn exceeds the depth of focus of the focused laser beam. In some micro-machining applications the system positioning requirements may be relaxed if certain tolerances are maintained, or if external hardware manipulates the device position, as might be done with a micro-positioning sub-system. The device may comprise a miniature part (e.g., single die) which is positioned by an external micro-positioning subsystem to a predetermined reference location. Similarly, if a miniature part has a pre-determined tolerance the positioning may be based on single measurement at a reference location or perhaps a single depth measurement combined with a lateral (X,Y) measurement. For processing of multilevel devices on wafers, (e.g.: 300 mm) at high speed it is expected that densely sampled three-dimensional information will improve performance, particularly as link dimensions shrink.

[0130] In applications requiring very high speed operation over a large surface (e.g., 300 mm wafer), an alternative method is to combine information which may be predetermined (e.g., the plane of a wafer chuck relative to a beam positioner plane of motion measured during a calibration process) with dimensional information obtained from each part to be processed. For example, in '534, FIGS. 1-2, a fraction of the tilt of region 28 may be associated with fixturing). For example, the steps may include (a) obtaining information identifying microstructures designated for removal, (b) measuring a first set of reference locations to obtain three-dimensional reference data, (c) generating a trajectory based on at least the three-dimensional reference data to obtain a prediction of beam waist and microstructure surface locations, (d) updating the prediction during relative motion based on updated position information, the updated position information obtained from a position sensor (e.g., encoder) and/or from data acquired during the relative motion. The additional data may be measurement data acquired at additional alignment target or at other locations suitable for an optical measurement (e.g., dynamic focus). Reference 2 describes a system wherein a precision wafer stage is used to position a wafer at high speed. A method of obtaining feedback information with resolution of a fraction of one nanometer is disclosed wherein interferometric encoders are used, and such a high precision method is preferred. In Reference 2 it was noted that other conventional laser interferometers may also be used. FIGS. 9-11 and columns 5-6 of Reference 2 describe aspects of the precision measurement subsystem associated with the precision positioning apparatus. Additionally, designated reference locations on the workpiece (e.g., wafer) which may be an x,y alignment target or a region suited for a three-dimensional measurement may be used for various applications. It should also be noted that height accuracy of about 0.1 μm was reported in “In-situ height correction for laser scanning of semiconductor wafers,” Nikoonhad et al., Optical Engineering, Vol. 34, No. 10, October 1995, wherein an optical position sensor obtained area averaged height data at high speeds. Similarly, a dynamic focus sensor (e.g., astigmatic systems used for optical disk tracking and control) may be used to obtain height information provided the data rate is fast enough to support “on the fly” measurement.

[0131] Various combinations of the above technologies can be used depending upon the application requirements. A combination may be based on the number and typical distribution over a device of microstructures designated for removal. When a large number of repair sites are distributed across a device, the throughput may be maximized by providing updates “on the fly.”

[0132] In an application of the invention, the target structure 10 is provided as a part of a multi-material, multi-layer structure (e.g., redundant memory device). The multi-layer stack having dielectric layers 14 , 15 provides spacing between the link and an underlying substrate 17 . In one type of multi-layer memory device, alternating layers of Silicon Dioxide 15 and Silicon Nitride 14 may be disposed between a copper link target structure 10 and a Silicon substrate 17 . The copper target structure is generally located in proximity to other similar structures to form a 1-D or 2-D array of fuses which are designated for removal. In addition to the copper link structure, underlying conductors 16 disposed as part of the functional device circuitry, may be in proximity to the link structure, and arranged in a series of patterns covered by relatively thin (<0.1 μm typical) Silicon Nitride 14 and thicker (1 μm typical) Silicon Dioxide 15 materials.

[0133] The irradiance distribution at the link may substantially conform to a diffraction limited, circular Gaussian profile. In another useful embodiment, the beam has an approximate elliptical Gaussian irradiance profile, as might be produced with an anamorphic optical system, or with a non-circular laser output beam. In one embodiment, the incident beam has a non-uniform aspect ratio 12 , 11 as also illustrated in FIG. 4 b (e.g., 3:1). Alternatively, rectangular or another chosen spatial profiles may be implemented in a lateral dimension. For example, Reference 1 discloses various advantageous methods and optical systems for “non-Gaussian” spatially shaping of laser beams for application to memory repair.

[0134] With the nearly diffraction limited elliptical Gaussian case, the preferable minimum beam waist dimension at location 11 approximates the narrow target 10 dimension of FIG. 1 b , which, in turn, produces high pulse energy density at the link. Further, with this approach, a high fraction of the laser energy is coupled to the link and background irradiance is reduced.

[0135] A typical copper link used in a present memory has width and thickness of about 1 μm or less, for example, 0.6 μm, and length of about five microns. Future memory requirements are expected to further reduce the scale of target dimensions. The minimum beam waist dimension Wyo at 11 will typically overfill the sub-micron link to some degree, whereas aspect ratio Wxo/Wyo 12 , 11 with Wxo a few microns along the link, can facilitate clean link removal. Additionally, rapidly decreasing energy density on the layers 14 , 15 and substrate 17 is achieved through defocus of the high numerical aperture beam portion 11 .

[0136] The graphs of FIGS. 5 a and 5 b illustrate the estimated defocus for various aspect ratios, relative to a circular Gaussian and an elliptical beam at best focus. FIG. 5 a shows the very rapid falloff of a 1.6 μm circular Gaussian (0.002 mm numerical divisions =2 μm). FIG. 5 b shows a normalized result to scale the energy density at best focus for the different spot shapes. These results indicate that with precision beam positioning in depth, wherein the power density is maximized at the target site, at relative reduction in energy density of more than one decade occurs at the substrate level for an exemplary multi-layer stack used in a copper based process for memory fabrication. Further, the rapid defocus relative to the waist WyO is beneficial for avoiding inner layer damage, provided the “tails” of the incident beam irradiate functional inner layer 16 (e.g., copper) at a low level.

[0137] In one embodiment for processing a multilevel device, copper link removal is initiated with application of the fast rise time pulse, having a nominal 10-90% rise time 4 in a preferred range of less than 1 nanosecond to about 2 nanoseconds. A pulse duration 5 in the range of about 2 nanoseconds to 10 nanoseconds is preferable to sever the link while limiting thermal diffusion. Pulse energies in the range of about 0.1 microjoules (μj) to 3 μj were effective, with a preferred typical range of about 0.1-5 /μj considered sufficient margin for spot shape and process variations. The preferred pulse duration may be selected based upon the nominal link thickness specifications, or based on a model of the dissimilar thermal and optical properties of adjacent materials. During the pulse duration, thermal shock of top layer 13 and thermal expansion of the target 10 result in explosion of the link through ruptured top oxide layers 13 , which in turn reduces the stress at the lower corner of the link structure adjacent to the layer 14 . The laser pulse is rapidly terminated, preferably within a few nanosecond fall time 6 after the explosion, at a time just after the thin link is cleanly severed, and prior to a time the lower corner of the link results in cracking of at least layer 14 . Further details and results related to the interaction of a laser pulse with a metal link and overlying layers is disclosed in references 4 and 5. The '471 patent and the associated specification describe the interaction process (e.g., FIGS. 1a , 1b, 11a, 11b, and in column 18).

[0138] Hence, a combination of the spatial characteristics (e.g., beam waist shape and position) and the temporal (e.g., rise time 4 , flatness, and duration 5 ) pulse characteristics avoids undesirable cracking of lower layers 14 , 15 , avoids significant pulse interaction with inner layer conductor 16 , and limits substrate 17 heating. Hence, despite the high reflectivity of the copper link at visible and near infrared wavelengths, and the expectation in the prior art of incomplete removal and damage to surrounding structures and substrate, the target structure is processed without undesirable damage to other structures. It is also known that copper, in addition to having nearly maximum reflectance in the near IR, is also more reflective than other link materials (e.g., aluminum, platinum). Nevertheless, due to the optical interaction of the near IR beam with the target and the optical and thermal properties of adjacent (overlying) layers, the preferred copper material can be processed.

[0139] Furthermore, near IR (Infrared) wavelengths also conveniently correspond to wavelengths where high bandwidth laser diodes are available, and to the spectral range where optical amplification of the pulsed laser beam can be efficiently produced with fiber and waveguide amplifiers. Those skilled in the art will recognize that amplified laser diode outputs, having a desired temporal pulse shape, may also be frequency multiplied to produce visible laser outputs when advantageous. The fast rise time of semiconductor diodes is particularly advantageous for producing a fast rise time, square pulse characteristic. Future developments in visible diode and optical amplifier technology may support direct pulse amplification in the visible range.

[0140] In a preferred system for copper link blowing, the link width is a fraction of one micron and the link spacing (pitch) is a few microns with present process technology. The link width may typically correspond to a wavelength of visible light. Further, at the microscopic scale of operation, where the lateral and/or thickness dimensions of the materials of FIGS. 1 b and 1 c are on the order of the laser wavelength, the thickness and indices of refraction of the stack materials can significantly affect the overall optical characteristics of the stack.

[0141] In one embodiment of the invention, a preferred reduced wavelength is selected in the visible or near infrared range wherein a non-absorptive optical property of the layers (e.g., interference or reflection loss) is exploited. The device structure of FIGS. 1 a and 1 b can be damaged with substantial absorption within the lower layers, such damage is prohibitive because of the presence of adjacent circuitry. This is in contrast to link processing with the prior art system of FIG. 2 b where inner layer damage is not generally detrimental to overall device performance.

[0142] U.S. Pat. No. 6,300,690 (Reference 8) describes a system and method for vaporizing a target structure on a substrate. The method includes providing a laser system configured to produce a laser output at the wavelength below an absorption edge of the substrate. Furthermore, Reference 4 discloses benefits of a wavelength less than 1.2 um for processing links on memory devices wherein the substrate is Silicon, namely smaller spot size and shorter laser pulsewidths. In accordance with the present invention, improved performance can be realized by exploiting the non-absorbing stack properties with wavelength selection. Furthermore, at least one of precision positioning of a high numerical aperture beam, spatial shaping of the spot, or temporal pulse shaping also will provide for reduced energy at the substrate. The result corresponds to a relatively low value of energy expected to be deposited in the substrate, despite an incident beam energy necessary to deposit unit energy in the target structure sufficient to vaporize the target structure.

[0143] The factors affecting the energy deposited in the substrate are, in effect, multiplicative. Likewise, at short visible wavelengths, copper is absorbing (e.g., about 50% at 500 nm, 70% in the near UV, compared to 2% at 1.064 um) so less energy is required for clean removal, at least an order of magnitude. The preferred identified wavelength corresponding to a relatively low value of the energy expected to be deposited in the substrate is within a visible of near IR region of the spectrum. A model-based approach may be used to estimate the shortest wavelength with sufficient margin for a specified dielectric stack, spot position, tolerance, temporal and three-dimensional spatial pulse characteristics.

[0144] For processing on links on multilevel devices with Silicon substrates, the limiting wavelength corresponding to a relatively low value of the energy expected to be deposited in the substrate (e.g., below the image threshold) may be within the green or near UV region of spectrum, but the use may require tightly controlled system parameters, including possible control of the stack layer thickness or index of refraction.

[0145] With wavelength selection in accordance with the present invention, where the internal transmission and preferably reflection of the stack is at or near a maximum, stack layer damage is avoided. Furthermore, decreasing substrate irradiance, while simultaneously providing a reduced spot size for link removal (at or near diffraction limit), is preferred provided irradiation of functional internal layers is within acceptable limits. Spectral transmission curves for typical large bandgap dielectric materials generally show that the transmission decreases somewhat at UV wavelengths. For example, in HANDBOOK OF LASER SCIENCE AND TECHNOLOGY, the transmission range of Silicon Dioxide is specified as wavelengths greater than 0.15 μm. The absorption coefficient of both Silicon Nitride and Silicon Dioxide remains relatively low in the visible range (>400 nm) and gradually increases in the UV range.

[0146] FIG. 3 is a graph which illustrates the estimated back reflection produced by a representative multi-layer stack of 14 Silicon Dioxide 15 and Silicon Nitride 14 pairs over a range of near IR wavelengths, where the thickness of the layers is about 1 μm and 0.07 microns, respectively. In accordance with the present invention, a large number of layers can be accommodated, and may range from about 4-28 dependent upon the process (e.g., sometimes multiple layers may separate a functional conductor layer).

[0147] By way of example, it is shown that significant reflection occurs over relatively broad wavelength range. A single layer disposed as an internal layer 14 will typically reflect roughly 2% at each surface over the visible and near IR spectrum. It is well known in the art of link and semiconductor processing that Silicon absorption varies by orders of magnitude in the near IR spectral range. Further, studies of Silicon material processing have shown that the absorption is unstable and non-linear with increased laser power and substrate heating at wavelengths near the absorption edge, as taught in reference 4. However, as stated above, the shorter wavelengths are preferred to produce smaller spots (references 4-6, and 8) and higher energy concentration at the link position.

[0148] In accordance with the present invention, exploiting the layer reflection with wavelength can further enhance the system performance and supplement the benefits associated with temporal and spatial control of the pulse in a preferred short wavelength range. Such wavelength selection is regarded as particularly advantageous at wavelengths where the substrate absorption would otherwise greatly increase, and significant margin can be obtained when the number of layers 14 , 15 disposed between the link and substrate substantially exceeds the number of overlying layers 13 . A preferred structure for processing will comprise a substantial number of layers, with large reflectance at a predetermined short wavelength, the wavelength being well matched for generation of the preferred fast square temporal pulse shape.

[0149] Standard laser wavelengths in the range of FIG. 3 include 1.047 μm and 1.064 μm, the latter being a standard wavelength of semiconductor diodes. Further, custom wavelengths include 1.08 μm, and other wavelengths generated with Raman shifting. Those skilled in the art will recognize that frequency multiplication of the near IR wavelengths can be used to generate short wavelengths, and with appropriate design multiple wavelengths may be provided in a single system. For instance a preferred temporal pulse shape, with a fast rise time, may be generated in the green portion of the visible spectrum by frequency doubling a near IR laser.

[0150] In an alternative embodiment, wavelength tuning is used to match the wavelength to the approximate peak reflectance of the stack. Such an arrangement may be particularly advantageous for adjustment of a laser wavelength at the edge of the reflectance range (i.e., “cutoff” range) over a limited wavelength range, whereby sensitivity to tolerances in the material thickness and index of refraction are avoided. As noted above, further discussion of laser amplifier systems and application to other link structures can be found in references 4-6.

[0151] Generation of the pulsed laser beam may include the step of shifting the wavelength of the laser beam from a first wavelength to a predetermined wavelength. The predetermined wavelength may be based on material characteristics comprising at least one of: (1) coupling characteristics of the microstructure, (2) multi-layer interference, and (3) substrate reflectivity.

[0152] Experimental results have shown that at a wavelength of 1.047 μm, where the absorption of Silicon in orders of magnitude higher than at 1.2 μm, substrate damage is avoided with a short q-switched (standard) pulse and the stack characteristic of FIG. 3 . However, the results with a standard laser having a q-switched temporal pulse shape showed cracking of an oxide layer 14 below the link. The relatively slow rising q-switched pulse shape, which for a Gaussian approximation is a substantial fraction of the duration, was considered a limiting factor for link removal without cracking of the inner layer based on experimental results. However, based on the teachings of the prior art, severe damage to the Silicon substrate would be expected at the 1.047 μm wavelength because the absorption is orders of magnitude higher than at a wavelength corresponding to maximum transmission. In accordance with the teachings of the present invention, the spatial pulse characteristics and the stack reflection are important factors to consider so as to avoid inner layer and substrate damage and short wavelengths of operation (which also provide for a smaller spot size and higher energy concentration at the link). Further, in accordance with the present invention, a predetermined square pulse shape generated at a laser wavelength of 1.047 μm would be expected to produce clean removal without undesirable changes to the stack and substrate.

[0153] Laser Processing and Process Design at the Sub-Micron Scale

[0154] Furthermore, in an exemplary advantageous embodiment for short wavelength processing of reflective microscopic structures, a specification for a multi-layer stack may be considered in process design. For example, a quarter-wave stack of alternating dielectrics or other suitable arrangement having a large difference in the index of refraction, and high transmission within each layer, is specified at a selected wavelength. It can be shown that very high reflectance is achievable, the quarter-wave stack being easily computed in closed form and modeled. Hence, the method and system of the present invention can be used effectively with other aspects of process design, and may be advantageous where the absorption of deeply buried layers and the substrate is relatively high, or where the width of a target structure is well below the laser wavelength.

[0155] The design of the device structure may have certain constraints related to the layout of the circuitry. As such, certain thickness and material for a certain layer may be defined, for instance an insulator in a plane of a conductor having the approximate thickness of the conductor, or related to the thickness of the conductor. It may be possible to select a material having a different index of refraction than the specified layer. A specified thickness may be based on the estimated reflection at an advantageous laser wavelength which may reduce or eliminate a requirement for special laser equipment operating at “exotic” wavelengths where the lasers are difficult to manufacture with high yield. The reflection may be estimated using a model wherein the thickness is a variable, and an estimate made to maximize the reflection, subject to other device constraints.

[0156] Thickness of the layers can be tuned to a wavelength in as much as the wavelength (or angle) can be tuned to the layers. Index of refraction could be used to fine-tune over a limited range, but the range may not be significant for small changes in index. Even with all thicknesses fixed by the process, the addition of a variable thickness tuning layer or layers with predetermined thickness could be used to significantly affect reflectivity of the whole stack. For example, a layer not constrained by metallization requirements could be used as a precision spacer between an upper and a lower stack portion. This could be a very powerful tool for tuning the process with adjustment of perhaps only one layer.

[0157] Physical Optics and Laser Processing of Multi-Level Devices

[0158] Other controllable laser characteristics may be exploited, alternatively or in conjunction with wavelength selection, to provide further improvements in the processing energy window. Reference 3 describes an advantageous method and system for polarization control, including dynamic polarization selection and computer control so as to align the polarization with a link orientation (e.g., details shown in FIG. 4 and the associated description in the reference). The polarization can be selected on the basis of the target coupling characteristics, the film reflectance, or a combination thereof.

[0159] With a link dimension below the spot size, effects like diffraction, scattering, and edge reflection should be considered as physical phenomena which can have either advantageous or detrimental results depending upon the device geometry and beam characteristics. Likewise, at high energy density, non-linear absorption may affect results, with particular concern of semiconductor material damage.

[0160] An additional important consideration with fine pitch (spacing) of adjacent links and circuitry is collateral damage. Furthermore, functional circuitry in a plane of the layers must not be damaged. With an increasing trend toward fine pitch and high density memory, the three-dimensional structure of the device should be considered and may affect a choice of beam spatial and temporal characteristics. By way of example, FIGS. 4 a - 4 c illustrate effects of reflection and diffraction associated with sub-micron width link 10 resulting in truncated 43 , 44 Gaussian beams 11 , where the spot size (as measured at the 13.5% point) is wider than the link by varying degrees. The sketches are representative of a diffraction limited beam waist at a near IR wavelengths. The central lobe is clipped by the link, which appears as a near field obscuration, resulting in transmitted beam portions which are truncated 43 , 44 . Energy which is not incident on the link may propagate at wide angles into the layers 49 which may be advantageous from the standpoint of avoiding damage to the substrate 17 as shown in FIG. 1 . In any case, there will be some correlation of neighbor irradiance with spot size. Large spots with a relatively large depth of focus have reduced divergence and neighbor irradiance can be small, provided that the link spacing is large enough that the non-absorbed energy 43 of the incident beam impinging on an adjacent structure is weak, for instance corresponding to level 44 . With a higher N.A. and smaller spot size, the reflected beam diameter at the link location 46 is increased. There will be a maximum value for some spot size 41 , 42 . Then irradiance at a neighboring link 48 decreases as the reflected energy grows larger in area.

[0161] Simultaneously there is an angular variation in internal reflection. Hence, the stack layer thickness can also effect the irradiance of adjacent structures, including the internal structures 16 of FIG. 1 . Furthermore, polarization variations with angle are expected to produce variations. FIGS. 6 a and 6 b illustrate by way of example geometric ray tracing effects of internal reflections propagating over an extended area.

[0162] Similarly, as shown in FIG. 4 c , if a portion 45 of the laser beam incident on the edge of the link 46 is considered, the energy which is not coupled into the link structure may also be scattered and/or specularly reflected to the adjacent links 48 . The inter-reflection 47 occurs as a consequence of at least the link 46 physical edge profiles which may be slightly curved or sloped.

[0163] An additional consideration is the three-dimensional spacing between an inner conductor layer 16 of FIG. 1 , the beam waist 11 , and the adjacent links 48 of FIG. 4 c . A large numerical aperture beam waist 11 , producing the smallest spot size at the link, while diverging and reflecting in a manner so as to avoid significant interaction with the inner layer 16 is preferred. Examination of FIGS. 4 a - 4 c suggest a reduced spot size with controlled precision 3D waist positioning is expected to reduce collateral damage by maximizing energy coupled into the link. With high enclosed energy within the link and a low intensity transmitted profile 44 , edge reflection is minimized. The spatial profile should also be selected subject to the constraint of only low level, negligible interaction between the beam angular distribution at 16 .

[0164] It is preferred that the interaction mechanisms associated with a portion of the three-dimensional device structure be modeled for selection of at least a spatial pulse characteristic, such a characteristic may be the N.A. and position of the beam waist. Preferably, the model will include an estimate of the irradiance seen by each adjacent link structures 48 , internal layer 16 , and substrate 17 . Whereas damage to adjacent link structures may be relatively apparent with conventional microscopy, assessment of inner layer 16 and substrate 17 damage may be considerably more difficult with the 3-D device structure.

[0165] With link widths below 1 μm, and pitch of a few microns, precise, sub-micron alignment is required to compensate for variations between wafers, and local variations within a wafer, and system tolerances (e.g., 300 mm wafer with 25 μm of topographical variation, and 5 μm of manufacturing tolerances, for instance). In accordance with the present invention, a precision positioning method and system is used to relatively position the beam waist so as to provide high laser energy concentration at the link. Also, one important consideration for precision positioning is predicting accurate (Xlink,Ylink) location information. The prediction is subsequently used by a motion control and positioning system to generate a laser output via trigger 2 at the target coordinates, during relative motion of the target 10 and laser beam. A preferred embodiment includes a polarization insensitive scanning and detection system as described hereinbelow, wherein a region containing an alignment target location is imaged to obtain reference data. The target location is often covered by a dielectric layer of Silicon Dioxide, Silicon Nitride, or other insulating material. Experiments have indicated that polarization insensitive detection is advantageous to avoid spurious measurements. The results led to a hypothesis that that birefringence is introduced in the insulating layers by polishing or other process operations, which is manifested by polarization variations in the reflected beam. These variations reduce the signal-to-noise ratio and appear to induce position distortion. The digital output data from each target location is used by an 8-parameter least squares alignment algorithm to estimate and correct position information affected by offset, angle, scale, orthogonality, and trapezoidal variations over the wafer containing the links to be processed.

[0166] Given the variations in the received beam at the target location, concerns arise that process variations may affect layer optical properties near the target structure. Furthermore, in practice, variations occur in the thickness and reflectivity of the target and layers, either over a wafer to be processed or from batch-to-batch. Measurement of the thickness and reflectivity is useful for process monitoring, and can also be used to determine adjustments for the laser power and wavelength to increase the energy window. For instance, any variation in the reflectivity of the link can affect the energy required for processing. A preferred method and system for adaptive energy control is also described hereinbelow.

[0167] As dimensions of links and other microscopic structures continue to rapidly shrink, those skilled in the art will appreciate the benefits of multi-parameter modeling. A model-based approach leads to selection and precision control of the spatial and temporal characteristics of the laser output, resulting i