Improving the state-of-the-art IC technology made it all possible.

• Price: Moore's Law stated that we could double the number of transistors we could integrate every two to three years. Its only purpose, to give us lower cost and any correlation to performance or power, no longer applies. In 2008 there was even talk of slowing integration rates because we're having trouble taking advantage of the capability we had. But even though we can now put hundreds of billions of transistors on a device and the cost of these devices are in the range of $1 per billion transistors, we now are comfortable with how to take full advantage of the technology.
• Performance: Increasing performance through multiprocessing is a concept that has been around for 40 years. In 2008, we predicted we'd see SoC devices that integrate sophisticated "heterogeneous multiprocessor" architectures incorporating many diverse processing blocks together, to relatively simple "homogeneous multicore" devices grouping cores similar in nature. Today, processors look like microsocieties with each core performing its own assigned task, perhaps unaware of the overall application. And, just as we predicted, devices have thousands of these blocks integrated into those microsocieties.
• Power: We achieved ultra-low power by moving to transistors that run at lower voltages, operate at lower frequencies, and power down when not in use. We now have predictive circuitry, which turns on circuits only when they're needed. Power scavenging was also a major breakthrough. Many embedded applications arose as we became comfortable with the concept of the perpetual device--a device that scavenges energy from its environment. With the perpetual device, we were able to implant medical devices with a lifetime source of energy and stress sensors buried in bridge support columns.
• Integration: We'd still be stuck with clunky 2008 products if we weren't able to simplify system-integration roadmaps with three aspects. The first step was to create the system at a board level with multiple devices on a printed circuit board--chips-on-board (COB) architecture. Secondly, we put the system in one package using stacked die or multichip packaging--a system-in-package (SIP). The ultimate integration was to put the system on one substrate--system-on-chip (SOC). But this SOC quickly became a "sub-system on a chip" and was integrated in as part of a larger system-in-package (SIP). Integration, it turned out, was solved not by SOC, but rather by the SIP.
• Development environments: Year 2028 processors have hundreds to thousands of cores, something almost unimaginable back in 2008 when the complexity of multiprocessing just four cores frustrated designers. Over the years, developers demanded tools to design multiprocessor systems without having to explicitly program each core. As we reluctantly predicted, we did achieve design nirvana whereby we have significantly reduced the design complexity of a sophisticated system. It is comparable to reducing the complexity of a multidimensional crossword puzzle to that of two-dimensional TicTacToe. Developers now rely on their development environments to manage all their lower-level expectations, including component configuration, integration, interfacing, and management. Development tools also determine what is efficient to execute in parallel and how to partition it.