The core of his argument is this:
... in the coming decades, researchers will have to do something entirely different – design a competently new kind of computer.
That is not a new. The current design is call von Newman.Others designs are possible. Neural Networks is very under explored.
Here are quotes from other sources. Nobody says there is only one way to make computers.
The von Neumann architecture is a design model for a stored-program digital computer that uses a processing unit and a single separate storage structure to .
In the original "Harvard computer", built in 1944 and for which the architecture is named, the program-handling task and the data-handling task were sufficiently different to result in two different storage technologies. Today, the vast majority of computers are von Neumann architecture because of the efficiencies gained in designing, implementing, and operating one memory system instead of two.
What all this stuff means is that if you break away from the prevalence design you can still call it a computer. No all computers have to follow that same fundamental design.
So a future computer may be so efficient that you can not really re-program it to do something it was not made to do. But it still could be a very useful device. In some respects, thah could be an advantage.
At one time we had Analog computers. And they were serious machines in a laboratory string.They were used to solve complex mathematical equations. And they where very fast.
Just not very precis. WAIT. Read this:
A VLSI analog computer/digital computer accelerator
Cowan, G.E.R.; Melville, R.C.; Tsividis, Y.P.
Solid-State Circuits, IEEE Journal of
Volume 41, Issue 1, Jan. 2006 Page(s): 42 - 53
Digital Object Identifier 10.1109/JSSC.2005.858618
Summary:The design of a single-chip VLSI analog computer fabricated in a 0.25-μm CMOS process is described. It contains 80 integrators, 336 other linear and nonlinear analog functional blocks, switches for their interconnection, and circuitry to enable the system's programing and control. The IC is controlled, programmed and measured by a PC via a data acquisition card. This arrangement has been used to simulate ordinary differential equations (ODEs), partial differential equations, and stochastic differential equations with moderate accuracy, significantly faster than a modern workstation. Techniques for using the digital computer to refine the solution from the analog computer are presented. Solutions from the analog computer have been used to accelerate a digital computer's solution of the periodic steady state of an ODE by more than 10×.
We are not yet down to the level of where a single molecule can repent a binary state of either a one or a zero.
The article in the topic was a was of time.IMO.