Learning from recording head manufacturing about 3-D structural process control
07/01/2003
The industry's trend to more device structures stacked on a die, increasingly including copper interconnects and low-k dielectrics, has forced the need to develop techniques that will let us see, measure, and ultimately control these structures. Conventional inspection and metrology techniques are approaching the limits of the resolution, but more important, they are fundamentally incapable of providing extensive 3-D information required to characterize new structures.
We only need to look at thin film recording-head manufacturing for a viable solution to 3-D metrology that is immediately applicable to IC interconnect structures, among other applications, that require the evaluation of complex geometrical or delicate structures that are difficult to evaluate from a top-down perspective (Fig. 1).
null
The data storage industry's success has been due in part to the early adoption of advanced 3-D structural process control techniques, specifically dual-beam structural process control, and realization that characterization of overall structure is more important than single-dimension evaluation (see "3-D metrology success" below).
Dual-beam structural process control uses a focused ion beam (FIB) of relatively massive ions to cut cross sections through targeted features at precisely controlled locations to expose subsurface structure. This can be done at any stage of a fabrication process. Then a scanning electron microscope (SEM) uses low-mass electrons to provide high-resolution images of the exposed cross-section surface without causing further surface erosion. Switching between cutting and imaging modes is nearly instantaneous. In use, the wafer surface is usually oriented perpendicular to the FIB axis.
null
An analyst can use the electron image to closely control the progress of the ion beam. Successive images, acquired as the cross-sectioning progresses and compiled into a video clip, reveal the complete structure in a powerfully intuitive manner. The combined beams also overcome the conventional top-down limitations to CD-SEMs. Advanced dual-beam systems now include automated unattended pattern recognition, cross-sectioning, image acquisition, and metrology. A typical analysis includes a variety of measurements on each cross section and provides a precise, quantitative structural characterization.
In contrast to the wafer sacrificed with SEM analysis, dual-beam analysis allows the return of analyzed wafers to production, significantly reducing their measurement costs. Several major semiconductor manufacturers are currently evaluating the viability of allowing the wafers to continue in the process.
Structural process control for ICs
The adoption and successful ramp of copper-based interconnects requires the fabrication of complex multilevel interconnects, including high-aspect-ratio vias free of subsurface voids. The detailed 3-D shapes of features within each layer, as well as the relational positioning of features across and through multiple layers, are critical to device performance.
More specifically, the introduction of copper as an interconnect material and plating as a deposition process has brought a whole new set of concerns to process engineers.
Barrier-seed layer continuity is critical, but impossible to evaluate top-down when it occurs within re-entrant features typical of a damascene process (Fig. 2). Plated copper, used to create plugs that connect one layer of metal to the next, is prone to voids. In subsequent thermal processing, voids tend to coalesce into larger voids that migrate upward.
Traditionally, such defects were not detected until electrical test at the end of the fabrication process. Dual-beam metrology permits close control of each process that makes up the module: interlayer dielectric deposition, photolithography, etch, barrier-seed layer deposition, copper fill, and planarization. Moreover, these processes all interact to control the overall performance of the interconnect.
Moving metrology and inspection upstream from end-of-line, or end-of-module, permits defects at any step to be corrected before they result in defective structures, and often in time to salvage devices by reworking them. Further, conventional manual cross-sectioning and SEM imaging to assess the quality of these features is not feasible because soft low-k dielectric materials can smear during the sample preparation. Dual-beam analysis eliminates smearing and provides immediate results that can be fed back to control the processing of following lots, or fed forward to fine-tune subsequent processing of the analyzed wafers. It permits a fundamental shift in the inspection and measurement strategy, moving from event-triggered defect response to scheduled process monitoring and control.
Shallow trench isolation (STI) is another excellent example of the importance of detailed 3-D structural analysis (Fig. 3). STI places isolating trenches between devices, and the need to minimize trench size makes their detailed geometry very important. Their depth must be controlled to within 10–20nm. The angle formed by the trench wall and trench floor must also be carefully controlled to ensure uniform and continuous oxide coverage. The detailed profile of the trench also affects the stresses induced in the underlying silicon and controls the magnitude of undesirable leakage currents.
null
Conclusion
Approximately 95% of thin film heads manufactured today are controlled by dual-beam-based structural process control. The process is highly automated, reducing the milling and data acquisition time to <2 min/cross section, during which time the system may make as many as 15 independent measurements. Three-sigma repeatability is on the order of 1nm. This has enabled the data storage industry to achieve higher productivity, better process control, and predictability. The semiconductor industry has many similar requirements for 3-D structural analysis.
The current effort to integrate multiple new technologies across the manufacturing process only increases the need for good visibility on process performance at every step. There has been a fundamental shift from 2-D to 3-D structures and conventional inspection and metrology techniques cannot supply the needed information. The results have already shown significant savings in time to data, reduced lot hold time, and real-time process control.
Anantha Sethuraman, Jason Donald, Craig Henry, FEI Co., Hillsboro, Oregon
For more information, contact Anantha Sethuraman, VP of the Semiconductor Business Unit, MicroElectronics Product Group at FEI Co., 1135 E. Arques Ave., Sunnyvale, CA 94085; e-mail [email protected].
3-D metrology success with thin film heads
Economically, the most important performance metric for recording heads is the width of the track created by the head. This, in turn, controls the density of the data written to the recording media. Data storage manufacturers measure their performance by improvements in areal density (gigabits/in.2), and their equivalent to Moore's Law is that since 1997 they have progressed from 1 to ~100 Gb/in.2, doubling areal density every 12 months. Heads with 100nm minimum feature sizes are in production.
Like ICs, thin film heads are built up layer by layer on a substrate. The exact geometries and relative positions of various elements of the head are critical to performance, including lateral offset between the read head and the write head, the longitudinal height of the GMR stripe, and the stripe's position relative to the throat of the write head (see figure). These complex 3-D relationships are impossible to evaluate from a simple top-down inspection and, unlike silicon wafers, these substrates cannot be cleaved manually.
Because wafer fabrication for heads and subsequent system assembly are geographically separated in the head industry, there are delays of several weeks between various manufacturing processes. Before the implementation of dual-beam inspection, structural problems introduced during wafer fabrication often went undetected until optical inspection after slider fabrication or magnetic testing after head gimbal assembly. By the time a problem was detected, there were several weeks and several million dollars worth of defective work in the production queue.
null
Today, using dual-beam metrology, process engineers can automatically cross-section head structures as they are fabricated, deriving a complete 3-D analysis, revealing problems as they occur, avoiding additional investment in defective product, and often permitting rework to salvage investments already made. This same information, fed forward and backward through the process, improves overall performance through a sequence of ever-narrowing process windows.