How Faster COVID-19 Research Is Being Made Possible by Secure Silicon


When Intel and Leidos set up a “trusted execution environment” to enable a widespread group of researchers to securely share and confidentially compute real-world data, it was no small achievement.

(Image: bestber via Adobe Stock)

(Image: bestber via Adobe Stock)

Viruses are slippery things. They adapt and change and sometimes surprise you with a new trick in the wild – a mutation, a rare side effect that they didn’t first produce in a lab setting. These surprises are part of why it’s so critical for medical researchers to collect and share real-world data about what a diverse patient population is experiencing right now, outside of a lab, so that the best treatment can be found quickly.

Historically, though, securely sharing and computing large datasets has been a clunky, arduous, and even impossible process. It makes collaborative medical research more difficult, and it slows down the medical community’s ability to respond to real-world data.

So when Intel and Leidos set up a “trusted execution environment” that enabled a widespread group of researchers to securely share and confidentially compute real-world data about COVID-19, it was no small achievement.

The Usual Way
“There are a multitude of challenges” to this type of research, says Chetan Paul, CTO of Leidos, an IT systems integrator and service provider.

As Paul explains, most patients have several providers: a dentist, an eye doctor, a general practitioner, a cardiologist, etc. Each physician will have their own system. Certain health records – like medical images or DNA data – are very large files. There are strong regulations protecting the privacy and restricting the portability of health data.

The first challenge is the administrative headache of getting point-to-point data sharing and data use agreements with each provider, Paul says.

“They then put in conditions that you have to identify data and ‘data can’t leave my environment, so you have to run your work in my environment, use data in my format as-is,'” he says.

Analyzing holistic health data is a knotty task for just one patient. Now multiply that times tens of thousands of test subjects needed just for a clinical trial.

It may be excruciating. But it’s necessary.

“To do your study, you have to get a complete picture of the life cycle of the patients and the population,” says Paul. “But the data silos that are created for [each] individual patient are segregated, and it’s a nearly impossible task to bring them all into the same centralized location.”

So instead of dragging data kicking and screaming from various locations in order to analyze it at one central place, says Paul, why not conduct analysis at all those locations and bring the results back to one secure central location?

“That would be giving them the assurance that your data is segregated, safe in your environment, and I am working back in a secure and safe fashion,” says Paul. “Instead of fighting against the problem, we took cues from the problem. And that’s where the Intel technology was a perfect fit.”

A Perfect Fit
Leidos built this multiparty analytics solution atop Intel Software Guard Extension (SGX) in the new third-generation Intel Xeon Scalable processor, code-named Ice Lake, which was officially released in April.

SGX secures data in-use – not just in-transit or in-storage. Developers can partition sensitive information into “trusted execution environments” (or “enclaves”), which are areas in-memory on the processor that only allow access by authorized code. The enclaves are isolated from the rest of the environment to ensure transmitted information is encrypted and can only be decoded once inside the enclave.

It removes the need to move the data, explains Chris Gough, Intel’s worldwide general manager of health and life sciences.

“So if you can take an algorithm, encrypt it in a secure container, send that algorithm or application to the endpoint where the data resides, and run it there, not only are you improving the security, but you’re also hitting on a lot of other benefits, such as you’re not needing to duplicate the data, which of course, increases the attack surface,” he says.

This isn’t just a benefit to the stewards of protected health data, he says, but also to the developers who want to protect their intellectual property while running computations on the data.

Is this homomorphic encryption? Not quite. But they both fall into the category of privacy-protecting machine learning.

“One attribute of homomorphic encryption as it exists today is that it’s quite computationally expensive,” says Chris Gough, Intel’s worldwide general manager of health and life sciences. “I think SGX is better prepared to meet a large number of more mainstream use cases today.”

The Ice Lake generation of Intel Xeon scalable processors is a major step forward for SGX, Gough says, because it brings SGX to a mainstream server and allows for larger enclave sizes.

“I think SGX in its previous instantiation really didn’t have room to shine,” says Gough. “Those data-rich analytics, AI use cases that really benefit from confidential computing, SGX, and federated architectures were constrained by the smaller enclave size that was available previously.”

This enclave size is the game-changing factor for use cases like Leidos and others in life sciences.

“There’s a reason that [the Leidos project] is happening now and not two years ago,” he says. “I think some segments of the projects [in health and life science] that have stalled or not started because of ‘interoperability concerns’ are not always [about] interoperability. I think it’s the data rights. It’s the governance. It’s the sensitivity around regulated data. It’s concerns around intellectual property of the software that’s running against that data.

“So now,” he continues, if a developer “can take their algorithms, encrypt it in a container, send that container directly into a trusted execution environment in someone else’s data center, and that algorithm … can run against that data in that trusted execution environment where neither the owner of the data can access or see that algorithm and the owner of the algorithm can’t see the data and the results can be sent back, that is a paradigm shift that enables a level of collaboration across leading researchers, leading health-care providers and biotech companies to collaborate in ways that were really just not possible before. And I think it really will also serve to accelerate the development of the adoption of AI across our industry.”

Helping the Cure
Both Leidos’ Paul and Intel’s Gough say they are grateful for the opportunity to support vaccine research in any way.

“The workforce from these [research] agencies, they have worked tirelessly,” says Paul, “and we have been, I’d say, privileged and fortunate to assist them in any possible way.”

Gough worked on Intel’s Pandemic Response Technology Initiative, which has already approved over 200 proposals for technological collaborations. 

“Already I can tell that will be probably the highlight of my career,” he says. 

Sara Peters is Senior Editor at Dark Reading and formerly the editor-in-chief of Enterprise Efficiency. Prior that she was senior editor for the Computer Security Institute, writing and speaking about virtualization, identity management, cybersecurity law, and a myriad … View Full Bio

Recommended Reading:

More Insights