Taking a first look at the CUDA code, there are a few basic things I can see that will need to be done. First, the addition of more particles, all of equal mass, and with even 3D distribution throughout a tall rectangular space (0.5m * 0.5m * 2m) such that the structure of a pillar can be represented. Secondly, all the particles will need to have a short-range repulsion to another much like they do now in the Nbody simulation (to produce a similar effect to that of collisions). Next, all particles will need to have a range-restricted attraction to each-other, just outside the range of repulsion, and approximately the distance between particles in their starting positions (to simulate the structural connections between particles of the pillar/dust clumping). These two forces will need to be large enough that they aren’t interfered heavily by gravity (or at least seem to be heavily interfered), but also no so large that changes in velocity will be so high as the particles move wildly as they switch between these two zones. Next, a single particle needs to exert a large “flash” repulsive force (for a very small period of time) at the beginning of the simulation such that it degrades exponentially over distance (a inverse of a localized gravity equation would do). The exception to the nBody being that it must effect ALL other particles, and then cease. All particles will additionally need to be affected by gravity, and will need to be limited so they cannot pass below the a 2D plane that is determinate of the bottom of the pillar (say, y axis of -1m?).