The early promises of DNA computing to deliver a massively parallel architecture well-suited to computationally hard problems have so far been largely unkept. Indeed, it is probably fair to say that only toy problems have been addressed experimentally. Recent experimental development on algorithmic self-assembly using DNA tiles seem to offer the most promising path toward a potentially useful application of the DNA computing concept. In this paper, we explore new geometries for algorithmic self-assembly, departing from those previously described in the literature. This enables us to carry out mathematical operations like binary multiplication or cyclic convolution product. We then show how to use the latter operation to implement an attack against the well-known public-key crypto system NTRU.