In this contribution, we investigate how cone support layouts for reducing thermal displacements in Selective Laser Melting (SLM) can be optimized. SLM is an additive manufacturing technique in which objects are created layer by layer by melting a metallic powder with a laser at specific places. The melted metal solidifies and fuses to the part of the object underneath it, after which a recoater deposits a new layer of powder, and the process repeats until the object is finished. While solidifying, the dissipation of heat causes thermal deformations. These can create undesirable displacements of the object, which in some cases even result in collisions between the recoater and the part of the object that has already been printed. Recoater collisions have to be avoided, because they can cause the print to fail, and can damage the manufacturing equipment. For reducing thermal displacements, cone supports are often added to the print. However, since the addition of these supports requires more use of material and longer printing times, the total volume of the cone supports is best kept to a minimum. Standard practice is to rely on engineering judgement to design the layout of the cone supports, but an automated design process could more reliably lead to well-performing designs, while also reducing design times and costs. We therefore propose to use topology optimization for finding layouts of cone supports with a minimum total volume that still keep the thermal deformation of the printed object within acceptable limits. The cone supports are projected onto the finite element grid that is used in the optimization by using the feature mapping / geometry projection method. For modelling the forces that cause the thermal deformations, the inherent strain method is used. A direct optimization approach is investigated, minimizing the volume with constraints on the displacements to avoid recoater collisions, as well as an indirect approach, minimizing the compliance and iteratively adapting a constraint on the volume with a bisection algorithm until a volume reduction is no longer possible since it would lead to displacements that are too large. The indirect method is less precise, but can lead to reduced calculation times. The resulting designs are validated for real-world applications by a collaborating company that works in the additive manufacturing industry (Materialise), and show that the proposed strategy can lead to improvements in cone support design.