I am referring to both the design, and the independent, and auditable manufacture of the CPU. It should be noted that such a CPU needn’t fully compete with modern ARM, Intel, AMD, etc. CPUs, but it would be an incredible boon to have a fully trustworthy piece of hardware, even if it is considerably lower in it’s strength. For specifics, let’s say a CPU that could run a lightweight Linux distro at a “tolerable” speed.
Creating the designs for the CPU, of course while still difficult, is, most likely, the most feesbile aspect – I presume it would “just” consist of writing the Verilog, or some other hardware description language to describe the CPU’s function. The manufacture, however, is a substantial obstacle. Modern photolithography is, quite litterally, at the very forefront of human technological creation. I am just hoping that turning back the clock perhaps 20 years on the technological complexity might reduce the barrier to entry.
There are RISC-V cores, whose designs have been published, which are capable of running a lightweight Linux distro, and even SBCs with them on. T-Head’s C906 on the Nezha board is an example.
RISC-V is probably the closest thing we have. But manufacturing is still a huge deal, as you said. Making anything with close to modern performance means dealing with microscopic things. I think it’s gonna stay difficult to do until 3d printers advance a lot. Maybe decades if not longer.
RISC-V is just an instruction set – same idea as x86. While it is, of course, important to also have an open instuction set, that is somewhat separate from this post’s intention. I am referring to the physical manufacture of semiconductors, RISC-V, or otherwise.
I just stumbled over this project:
deleted by creator
While it may be RISC-V, that doesn’t exactly satisfy this post’s inquiriy. A CPU that uses the RISC-V instruction set may still have malicious hardware contained within it – think of the Intel managment Engine, for example.
deleted by creator
It’s all a question of how much money you have access to. OpenCores has a bunch of HDL processors that can serve as a base design that could be turned into an ASIC and sent to a fab company. From what I’ve heard it costs on the order of a million USD for a run of wafers. You will own the design fully, there’s very little chance they would insert anything onto the wafer you didn’t ask for. Difficult to truly verify though: you hand the vendor instructions like “put a D flip flop here” and they turn that into all of the process step lithography masks. You can shop that same design to multiple fab vendors but due to IP differences it might require tweaking per vendor.
Maybe if you work with a university’s clean room you could do it for less, but they won’t have a library of drop-in IP blocks for transistors, clock generators, PLLs, high speed I/O (eg pci lanes), memory, etc so now you have to develop those from scratch.
Similar concept if you were able to acquire some used equipment: all the process engineering needed to make a functioning transistor now falls on your project, as does sourcing the chemicals, getting the masks fabricated, handing the hazardous waste, maintaining a clean room, the list goes on.
All these hurdles are achievable of course, but that requires some serious engineering skills, and those folks tend to cost a lot of money.
The short cut to all of this is, if you aren’t trying to sell millions of these things, just implement the CPU in an fpga and call it a day. The newest mid-range FPGAs (~ $1k range) can readily reach 500MHz and have 8ish lanes of PCIe 5.0. The security mechanisms can also be used to prevent any other firmware other than yours from running on the chip (encrypted firmware plus an authentication layer).
deleted by creator
