We did make larger squares and bars in the past few years for internal prototyping. The issue is that as you increase in size, with a fixed number of capacitive channels (limited by the chip), you lose in spatial resolution, which is what made us settle for the current dimensions. Nothing prevents you from tiling several squares or bars or hex side by side to make a larger sensing surface. If you use them as a single sensor, the "centroid" reading from each of them will not be very usable, and you'd be better off reading raw data instead and do the centroid detection on the host device instead.