There is an asymptote that is being approached with near current feature size technology. Future optimizations will be on both size and power, and certainly much more parallelism and specialization, wringing every last bit. Certainly far more than double there. Yet research will as always provide some more marginal breakthroughs. And quantum understanding will initiate an entirely new race if that ever happens. Even if not, as physics and invention curve is reached, die size will go up until sustainable defect rates govern. And then that's it, for a very long time. A great leveling in ubiquitous compute power will occur. All will have access to the best, cheaply, as a boring commodity. Freely solar and wind powered even. Some say that is a good thing.