For years, hackers have focused on finding bugs in computer software that give them unauthorized access to computer systems, but now there's another way to break in: Hack the microprocessor.

On Tuesday, researchers at the University of Illinois at Urbana-Champaign demonstrated how they altered a computer chip to grant attackers back-door access to a computer. It would take a lot of work to make this attack succeed in the real world, but it would be virtually undetectable.

To launch its attack, the team used a special programmable processor running the Linux operating system. The chip was programmed to inject malicious firmware into the chip's memory, which then allows an attacker to log into the machine as if he were a legitimate user. To reprogram the chip, researchers needed to alter only a tiny fraction of the processor circuits. They changed 1,341 logic gates on a chip that has more than 1 million of these gates in total, said Samuel King, an assistant professor in the university's computer science department.

"This is like the ultimate back door," said King. "There were no software bugs exploited."

King demonstrated the attack on Tuesday at the Usenix Workshop on Large-Scale Exploits and Emergent Threats, a conference for security researchers held in San Francisco.

His team was able to add the back door by reprogramming a small number of the circuits on a LEON processor running the Linux operating system. These programmable chips are based on the same Sparc design that is used in Sun's midrange and high-end servers. They are not widely used, but have been deployed in systems used by the International Space Station.

In order to hack into the system, King first sent it a specially crafted network packet that instructed the processor to launch the malicious firmware. Then, using a special login password, King was able to gain access to the Linux system. "From the software's perspective, the packet gets dropped... and yet I have full and complete access to this underlying system that I just compromised," King said.

The researchers are now working on tools that could help detect such a malicious processor, but there's a big problem facing criminals who would try to reproduce this type of attack in the real world. How do you get a malicious CPU onto someone's machine?

The IDG News Service is a Network World affiliate.