Robert Baruch has a bunch of great videos and if you look at the website for his "Project 54/74" [1] he spends lots of time taking high resolution photos of decapped processors. Really, it's an amazing educational experience. If you guys can spare just a few dollars/pounds/whatever a month he has a Patreon page [2] where you can donate to help him continue his research.
I keep having an odd thought. What if each register was paired with an ALU? Much more circuitry, but you'd only need two busses instead of three - one for an operand and one for the result. The other operand would be local to the ALU. Interconnect would have to be exceptionally expensive for this to make any sense, but I am amused at the idea anyway.
Although neat in its own right, anyone whose looking at this project out of concern for subversion should know there's still process nodes in use that you can see with your own eyes. Well, you dissolve coating around random chips to put them under a microscope. You're limited to 350nm and up if I'm right that the property ended around 250nm. There were also 500nm and 700nm available when I last looked at shuttle runs.
If one wanted to go about and take the easy way out and instead of using discrete chips going with an FPGA. Which FPGA-chip / board would be a suitable option?
At the very top end, the VC707 ($3500) will let you run a quad core 64 bit RISC-V at about 100 MHz. At the bottom end you'd be looking at one of the boards supported by PicoRV32 (https://github.com/cliffordwolf/picorv32), and some of those are as little as $100. If you want flashing lights as in the video then the Nexys 4 DDR might be a good compromise (but you'd have to program the lights to flash yourself).
The very quick explanation is that when you have lots of long wires (and especially wires of different lengths) between the components of your chip, you have to wait for all the signals on a bus to arrive before you can latch them into the component at the next clock edge, and this slows down the achievable clock speeds.
To explain this properly would require explaining how FPGAs work and how they differ from ASICs. There's a rather good talk by Uli Drepper which covers some FPGA basics: https://www.youtube.com/watch?v=Q-SkioVSZWw
In any case 100 MHz for an FPGA processor design is not slow for an FPGA, nor is it something that can be solved with more money (bigger and smaller FPGAs suffer the same effect). For GHz designs you have to implement an ASIC. For comparison, SiFive's FPGA implementation runs at 100 MHz but their ASIC of near-exactly the same design runs at 1.5 GHz.
Yes, because the V in RISC-V is the Roman numeral five. See the Introduction chapter of https://content.riscv.org/wp-content/uploads/2017/05/riscv-s... (the very first sentence of that chapter, in fact, mentions the pronounciation: "RISC-V (pronounced “risk-five”) is a new instruction set architecture (ISA) [...]").
1. https://project5474.org/index.php?title=Main_Page
2. https://www.patreon.com/user?u=3190442