1
\$\begingroup\$

My department of the company I am working at is in charge of developing a circuit for production, and security in the embedded system is a crucial requirement. The current prototype of our embedded system runs Linux on a micro SD card (with a MPU, Am335x). My job right now is to change the design into something more secure -- something that can't be easily modified by the user. However, I haven't done this in the past so I am not really sure what are the guidelines.

My thoughts so far:

Look for an EMCC chip on a BGA package with an internal layer on the PCB for the traces. That way it would be impossible to reprogram unless the user destroys the PCB. However, I am having difficulties finding an emmc chip on Digikey, apparently they are not very common and they are all sold out.

Other thing I was thinking was not using an EMMC at all but a secure NOR flash. However, I am not really sure if that would increase the development time in my department because I suppose that would require modifying the device trees on our AM335x so that it works with a secure NOR instead of an EMMC or a micro SD card.

Edit: The security requirement I am referring to on this post is related to hardware only. We are using a TPM for secure boot and a remote attestation protocol to assert the state of the PCRs remotely. All software (new firmware) must be code signed to run. Also, all sensitive data in flash is encrypted. Thus I am not worried about someone desoldering and reading the EMMC. I am concerned however, about all the communication that goes from the EMMC to the AM335x, so I need to make sure it would be hard for someone to connect a test probe to the PCB traces. And of course, reflashing the chip and/or resoldering a new one. Those are the two basic requirements I am in need of.

Any guidance on this would be greatly appreciated.

Thanks in advance.

\$\endgroup\$
11
  • 3
    \$\begingroup\$ Giving the security design to someone without security background is destined to fail. Anyway, "something more secure" is not a security requirement. You start the security design by defining the threats you want to protect from. \$\endgroup\$ – Eugene Sh. May 30 '18 at 21:29
  • \$\begingroup\$ Eugene, although I appreciate your honesty, that's not a very helpful comment. When I say "More secure" I am referring to basically avoiding the following: probing the pins, and re-flashing the memory. Our design includes a TPM, and we use a remote attestation protocol to assert the states of the PCRs. Now it's basically related to the hardware side of security (software security is one thing, and that's taken care of), basically we are left with hardware-based security. \$\endgroup\$ – Luis Cruz May 30 '18 at 21:37
  • \$\begingroup\$ HW or SW based security are not very different in this aspect. Security is starting with definition of the threats. What you want to protect from? Rewriting the flash? Resoldering it with a different flash? Tapping with probes and reading the data? Desoldering and reading? \$\endgroup\$ – Eugene Sh. May 30 '18 at 21:40
  • 4
    \$\begingroup\$ And my first suggestion is not sarcastic or condescending as it might seem, but it is a sad fact acceptable as an axiom in cybersecurity domain (which I am working in by the way). \$\endgroup\$ – Eugene Sh. May 30 '18 at 21:46
  • 1
    \$\begingroup\$ If you are concerned with "changing that by hand" by someone in a garage, then your concerns are unfounded, it would be impossible to gather any useful information and re-engineer the protocol. But for a seasoned professional there is no big problem to desolder the eMMC, put some interposer or socket, and connect anything to the interface, even if you encapsulate everything into epoxy. So I am afraid you need to really access potential risks of breaching relative to cost of your efforts. Is your device really worth the effort? \$\endgroup\$ – Ale..chenski May 30 '18 at 23:23
3
\$\begingroup\$

If the attacker has physical access to your device, you've basically already lost. A state-level attacker can do whatever they want. A well-off corporation won’t be too far behind. Even a well-equipped hobbyist with basic tools at home can do a lot.

If you have a vulnerability in which physical access to the path between your storage device and processor compromises your design, I argue that you need to try and change to a device and protocol that transmits encrypted data over the wire. I'd assume storing encrypted images on eMMC would be decrypted in memory on the processor, so perhaps capturing SD traffic would not be the biggest deal, but your decryption key has to be stored somewhere secure.

\$\endgroup\$
0
\$\begingroup\$

Cover everything with epoxy. The more, the better. Of course it won't stop a determined adversary, but will just make it more difficult.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.