Indeed, the evolution will probably in the other direction, with the on chip network adopting algorithms from wider networks to deal with scaling problems. DRAM interfaces use to be pretty simple, now they are being trained almost like a DSL line.