The Need for Speed

We’ve previously identified the convergence occurring between surface mount technologies (SMT), used to connect packaged semiconductor devices on printed circuit boards, and advanced packaging (AP) technologies, in which connections between the semiconductor devices and to the outside world are incorporated in the packaging process using front-end-like, wafer-or panel-based manufacturing processes. What implications does this convergence have for inspection and metrology techniques used to control the process, ensure yields and maximize profitability?

The requirements for inspection and metrology are controlled primarily by the size, shape, and number of features. Here, we are looking at three-dimensional features that range from 10µm to 100µm size and may number in the tens of millions per wafer. Above all, inspection and metrology systems must have the basic measurement capability — resolution, sensitivity, accuracy and precision — needed to provide meaningful measurements. Once those criteria are met, they must have the speed to keep up with the manufacturing process — typically 25 to 50 300mm wafers per hour. All else equal, more measurement per unit of time means lower cost per measurement and potentially greater profitability for the process as a whole.

Surface mount technologies have long used phase shift profilometry (PSP) for inspection and metrology. PSP works by projecting a pattern onto a surface and viewing that pattern from some angle off the projection axis. Imagine a pattern of straight, regularly spaced stripes projected onto a flat surface from directly above. Viewed from an angle off the projection axis, the stripes remain straight, though the spacing will decrease as the angle increases. Put some bumps and depressions in the surface and the stripes will curve with the direction of the curvature distinguishing bumps from depressions. Now imagine the surface is not flat, but contains a number of vertical-sided, flat-topped features. The stripes in the pattern will be discontinuous and offset where they cross such a feature. Knowing the viewing angle and the offset, we can calculate the height of the top of the feature. By placing multiple cameras around the viewing axis, we can see and measure all sides of the features.

The “shift” in phase shift profilometry refers to shifting the phase of the projected sinusoidal intensity pattern multiple times to solve for the 3 unknowns: reflectance, fringe contrast, and phase. Typical systems shift the phase in increments of either 90 or 120 degrees. By projecting stripes that vary sinusoidally in intensity across the stripe, we can readily resolve phase changes of less than 5 milliradians, giving the technique very high sensitivity to changes in feature height. One problem PSP must overcome is ambiguity in discontinuous shifts. Each stripe looks exactly like every other stripe. With a single measurement it is not possible to know whether a shift is a fraction of a period or a fraction plus multiple periods, or even to know the direction of the shift. This problem can be solved by projecting different patterns with different frequencies and orientations. Sophisticated “unwrapping” software can integrate image data from multiple patterns and multiple cameras to generate a precise three-dimensional map of the inspected surface.

The resolution of an optical system is ultimately limited by diffraction as a function of the wavelength of the light the system uses. Visible light ranges from 400 to 700 nanometers, allowing the best microscopes to achieve resolutions down to about half a micrometer.In addition, the resolving power and light gathering ability of an optical system is also a function of a parameter optical designers call the numerical aperture. Higher numerical aperture lenses, which have higher resolving power, generally tend to have larger diameters and are more closely spaced to the object. Needless to say, designing a system that puts multiple cameras large enough and close enough to the object to achieve diffraction-limited resolution, while also preserving a field of view large enough to provide reasonable throughput, is a non-trivial exercise. CyberOptics’ highest resolution Multi-Reflection Suppression (MRS) sensors offer 3µm lateral resolution and 50nm vertical resolution over a 15mm x 15mm field of view. And the technology can scale to 1.5µm lateral and 25nm vertical resolution, nicely matching the dimensional “sweet spot” for current and next generation advanced packaging processes.

Given the ability to meet the basic measurement requirements of the process, the next consideration becomes speed and throughput. Perhaps the greatest speed advantage of PSP comes from it use of image data, gathered simultaneously for all points in the field of view, as contrasted to scanning techniques that acquire data sequentially, point by point or line by line. In addition, our use of multiple cameras allows us to gather data several times faster than other PSP designs that use a single camera and multiple projectors. Finally, the ability to acquire both 2D and 3D information in a single pass, further boosts measurement throughput. We can typically acquire 25 million 3D data points from a full field of view in about 150 msec. Data processing times, especially for the sophisticated image analysis and unwrapping routines used by PSP, have been greatly accelerated by rapid developments in graphic processing units (GPU), driven primarily by virtual reality and gaming applications. With data acquisition rates as high as 75 million 3D data points per second, we can inspect 100% of the surface of as many as 25 300mm wafers per hour.

CyberOptics has achieved great success recently, placing our MRS technology at a number of major manufacturers. We believe the reason for that success is the combination of measurement capability with a significant speed advantage. We expect PSP systems to become the measurement and inspection technology of choice for “middle-end” applications in advanced packaging.