MIT Researchers Develop A Tiny Chip To Safeguard Data And Enable Efficient Computing

Today’s digital world enables users to leverage smartphones to carry out various functions and services for a range of purposes.

This includes using complicated computations to gather insights in real-time for applications such as health monitoring, navigation and more.

These applications are powered by vast machine learning (ML) and artificial intelligence (AI) models, which require data to be moved between smartphones and a central memory server, which could result in slow and energy-inefficient performances.

To offset such issues, engineers use hardware that cuts the need of back-and-forth data movement. However, this could also result in systems that are vulnerable to attackers stealing secret information.

Now, researchers from MIT and MIT-IBM Watson AI Lab have come up with an announcement that’s set to reduce such vulnerabilities and bring in other performance benefits.

So, what did the researchers say? Let’s explore!
 

What Did The Researchers Announce?

 

• In a news release published on its website, MIT AKA the Massachusetts Institute of Technology, revealed that researchers from MIT and the MIT-IBM Watson AI Lab created a machine-learning accelerator that is resistant to the two most common types of attacks.

• As per the release, the new chip “can efficiently accelerate machine-learning workloads on edge devices like smartphones while protecting sensitive user data from two common types of attacks — side-channel attacks and bus-probing attacks.”

• This includes ensuring data like a user’s health records, financial information or other sensitive data are kept private.

• The researchers developed several optimizations that brought in stronger security without affecting output accuracy, however, it did result in slightly slowed device performance.

• Ahead of this, the team found that the chip could make devices slightly more expensive and reduce energy efficiency but could benefit AI applications such as augmented reality (AR), virtual reality (VR) or autonomous mobility.

• Despite its reduced performance, lead author Maitreyi Ashok, an electrical engineering and computer science (EECS) graduate student at MIT and lead author, believes its security enhancements make it worthwhile.

 

How Did The Researchers Achieve This?

 

• The team targeted digital IMC (in-memory compute) chips which are broken into pieces that store parts of ML models and reduce back and forth data movement. This was accompanied by a “three-pronged approach” to thwart side-channel and bus-probing attacks.

• The first move included effectively splitting data in the IMC into random pieces, which wouldn’t allow side-channel attackers to reconstruct data.

• The second included the use of a simple and lightweight cipher to encrypt the model stored in off-chip memory that decrypts pieces only when necessary.

• Third, the team generated a unique decryption key using random variations introduced during manufacturing, that decrypts the cipher directly on the chip.

• “Maybe one wire is going to be a little bit thicker than another. We can use these variations to get zeros and ones out of a circuit. For every chip, we can get a random key that should be consistent because these random properties shouldn’t change significantly over time,” said Maitreyi Ashok.

 

What Did The Researchers Say?

 

• Maitreyi Ashok, said, “It is important to design with security in mind from the ground up. If you are trying to add even a minimal amount of security after a system has been designed, it is prohibitively expensive. We were able to effectively balance a lot of these tradeoffs during the design phase.”

• Anantha P. Chandrakasan, the Dean of MIT’s School of Engineering, MIT’s Chief Innovation and Strategy Officer and co-chair of the MIT–IBM Watson AI Lab. said, “As security has become a critical issue in the design of edge devices, there is a need to develop a complete system stack focusing on secure operation.”

• [Contd.] “This work focuses on security for machine-learning workloads and describes a digital processor that uses cross-cutting optimization. It incorporates encrypted data access between memory and processor, approaches to preventing side-channel attacks using randomization, and exploiting variability to generate unique codes. Such designs are going to be critical in future mobile devices.”

While the additional security reduced energy efficiency and required a larger chip area resulting in increased fabrication costs, the team is planning to explore more methods to tackle these problems.

Maitreyi Ashok believes, “As it becomes too expensive, it becomes harder to convince someone that security is critical. Future work could explore these tradeoffs. Maybe we could make it a little less secure but easier to implement and less expensive.”

Do you think technology and AI companies should consider using more expensive chips to ensure security or should they consider cost-efficiencies above security?

Let us know in the comments below!