[Cryptography] ORWL - The First Open Source, Physically Secure Computer

[Jerry Leichter]

> 
> 1) Where are the open fabs where I can get a trusted processor design
> fabricated, and at reasonable cost?
> 2) Given the effectiveness of incredibly simple malicious hardware
> additions, some of which are difficult to notice even with a careful
> destructive analysis of the fabricated part, how can I verify that the
> fabricated design is indeed what I expected and consistently so?
> 
> I wonder if one could build a low-performance, high-cost but secure processor from an FPGA?  It would seem on the surface to be more difficult to embed an exploit in an un-programmed FPGA, especially if it had a very regular structure.
It's not the FPGA or anything else in particular that's the issue.  There are open-sourced versions of several chips around.  "Open" fabs are another issue.  But assuming that could be solved ... the problem is that the gap between the performance of proprietary silicon and what you could conceivably build using an open-source technology is immense.

On the one hand, you could probably build something today that would beat anything available 15 years ago.  On the other ... what we demand today has move substantially past that point.

Over the years, we've repeatedly come back to the idea of a "reference monitor" or "security kernel" through which all security-relevant decisions would flow.  Of course, as we've learned along the way, there are often all kinds of way to route around such central decision/control points.  But it seems as if the way to get at the issues here might be to have a (limited performance) open-source verified "security kernel" managing access to a high-performance unverified, possibly malicious, proprietary part.  Of course, the security properties the combined system is supposed to provide would have to be carefully specified and themselves verified.

It's not as if we haven't built this at the software level many times.  OS's run malicious user processes; hypervisors run malicious OS's.  Sure, there are bugs, but in principle all this stuff could be verified.

What would a hardware architecture analogous to this look like?  I'm sure such things have been done in the past.
                                                        -- Jerry


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.metzdowd.com/pipermail/cryptography/attachments/20160829/39abcced/attachment.html>