Photo: MIT
Engineers at the Massachusetts Institute of Technology (MIT) in the United States have made a recently developed type of security chip to ensure smartphone data protection.
The ‘MIT-IBM Watson AI Lab’ helped them develop this new type of chip, which is designed keeping in mind the security of smartphone data.
This breakthrough innovation is particularly relevant for various health-monitoring apps and mobile applications dealing with sensitive data, according to a report by science-based site Norridge.
The new A-chip also ensures that top AI models work efficiently across devices while preventing various cyber security threats.
Today, modern smartphone apps rely on various complex machine-learning models for health monitoring or money management.
Usually running such a model directly on a smartphone is a bit challenging. This is because of their size and constant data exchange with central servers, which slows them down and increases power demand.
In addition, continuous data transfer also increases the risk of cyber attacks.
To address such risks, a research team led by Maitreyi Ashok, a graduate student in MIT’s Department of Electrical Engineering and Computer Science, has developed a type of ‘machine-learning accelerator’ chip that builds strong resistance to common cyber attacks while maintaining the privacy of user data.
The chip is particularly useful in AI applications such as virtual reality or autonomous-driving, where the demand for computational power and safety is high.
According to Ashok, the inclusion of this technology may increase the cost of the device but also reduce power consumption. And, in terms of security, this cost can be called reasonable.
“Our aim was to ensure that security measures were built into the system from the outset, adding them later would have increased the cost,” explains Ashok.
Besides, he emphasized on working on the security measures of the new chip from the time of its design to handle various tasks in an efficient manner.
This research paper will be presented at the ‘iTriple Custom Integrated Circuits Conference’ to be held in Denver, Colorado. It will also highlight how the chip uses ‘Digital In-Memory Compute (IMC)’.
This process will facilitate various mathematical calculations in the device’s memory, which will be stored in different parts of the machine-learning model. This will reduce the size of data transferred and the potential for potential security breaches, Norridge wrote in the report.
‘IMC’ chips are generally vulnerable to cyber threats. So Ashok and his research team adopted a three-pronged strategy to increase chip security.
First, they developed a novel security system that splits data into different parts. This makes it difficult for attackers to coordinate, thereby reducing the risk of so-called ‘side-channel’ attacks.
Second, to protect against ‘bus-probing’ attacks, where hackers try to steal data directly from the communication system, the researchers used a lightweight encryption method, which protects data stored outside the main chip. This encryption system can be decrypted only on demand.
Third, the research team has developed a variety of encryption keys that reside directly on the chip. For this they followed a method called ‘Physically Unclonable Functions’.
This method uses variations that occur every minute in the chip’s manufacturing process to create a unique key, which strengthens the chip’s security without complex math.
Despite such success, these new security systems may increase power requirements and chip size, which may also impact manufacturing costs.
“In the future we will look at balancing security with chip manufacturing and power consumption. A little compromise in this regard can lead to a more realistic solution in the security system,” says Ashok.
