Question

I have a new computational architecture idea rather than Von-neuman. Current machines are based on Von-neuman’s CPU-Memory logical separation based. While in my case it is possible to fuse Memory-ALU to have a radically new CPU. In that case how i can compare my new architectural idea with existings?

• One way might be running various algorithm and compare. But unfortunately we haven’t that hardware yet.

• Emulation is okay? But it takes big effort to build a emulator from scratch

• How industry do compare?

Thanks in advance.

Answer 1

This idea has been explored to some extent in the past, for example here:

• The Berkeley Intelligent RAM (IRAM) Project
• Computational RAM at the University of Toronto
• The architecture of the DIVA processing-in-memory chip (PDF)

One of the problems with this approach is that CPUs and memory use different process technologies. If you put both on the same die you will have to compromise one of the two. A possible solution are 3D architectures where different dies are mounted one above the other.

A simulator (not an emulator!) would be an excellent way to compare your architecture against more conventional architectures. The closer the simulator is to a realistic hardware implementation the better.

If you want to test more than toy algorithms on your simulator, you will need a toolchain to compile regular software (say c++, java) on it. This is a mayor effort and probably even more complex than the simulator alone.

This is a project that could easily take several years for one person alone.

If you just want some feedback to your idea you could submit a paper to an academic conference and carefully read the comments of the reviewers. (This is what I would recommend.)