Trust never sleeps: Why hardware roots of trust are essential for security

Securing the enterprise is more challenging than ever. The number and sophistication of attacks have grown exponentially, while the defenses against them have not.

You can train employees to identify phishing emails, stop opening suspicious attachments, and use more sophisticated passwords. You can implement two-factor authentication, install intrusion prevention systems, and use artificial intelligence to detect malware. You can implement strict policies on data access and transport. But if the hardware you're using has been corrupted, none of these mitigation measures matter.

The problem of tainted hardware has been exacerbated by the ongoing pandemic. Supply chain disruptions have caused manufacturers to seek out lower tier vendors, increasing the risk of corrupted or counterfeit components, notes Bentsi Ben-Atar, chief marketing officer and co-founder of Sepio Systems, which makes zero trust hardware solutions.

Then there are the relatively new threats like Spectre and Meltdown, which attack vulnerabilities within CPUs. (When the security threats start to sound like Bond villains, you know things have gotten really bad.)

How can enterprises make the hardware that runs their networks and applications more secure? It starts with trust—or, more accurately, the lack thereof.

What is zero trust architecture, and why is it necessary?

Until recently, most organizations spent their security budgets on hardening the network perimeter. But once an attacker got past the initial authentication process, usually through brute-force attacks or social engineering (like phishing emails), they were free to move about behind the enterprise firewall. Many attackers have been able to lurk for months, collecting information about the company, its customers, and its employees.

In a zero trust framework, every user and device is continually validated and monitored, with limited access to data and network resources. Devices, applications, and data are also highly segmented, so if one part of the network is breached, attackers cannot easily access any other elements.

The Biden administration's May 2021 Executive Order on Improving the Nation's Cybersecurity calls for the use of zero trust architecture in both public and private sector networks. A fundamental element of secure architecture is validating the bona fides of every piece of hardware via what is called a root of trust.

What is a hardware root of trust?

Simply put, a hardware root of trust is a way to ensure the identity and authenticity of silicon devices at an atomic level.

Every semiconductor has a molecular structure as unique as a fingerprint or a snowflake. Even chips produced at the same factory, from the same wafer and at the same time, will vary ever so slightly from those on either side of them, notes Pim Tuyls, CEO and founder of Intrinsic ID, which sells root-of-trust IP to chip makers. Those variations produce infinitesimal differences in the amount of voltage required for each transistor to conduct electricity.

Please read: Constant scrutiny is the key to making zero trust happen

The differences in voltage can be used to produce a cryptographic signature (or root key) for the chip that never changes and cannot be cloned. Because the root key can be reproduced whenever it's needed, it's never stored, so it can't be copied or stolen. This technology is known as physically unclonable function (PUF).

When a device is powered on, initial boot code inside it looks for the appropriate key before it allows any code to run. Once the hardware and the rest of the boot code is authenticated, it can be trusted to load the OS and applications.

Tuyls likens the process to passing through an airport immigration checkpoint. Once an agent has scanned your passport, checked your fingerprints, and verified that you're not on a watch list, you are free to travel. Likewise, once a chip's root key and boot code have been validated, it can be trusted to run code and communicate with other devices.

Why do we need multiple roots of trust?

Other methods for implementing root of trust have involved burning read-only encryption keys into silicon during the manufacturing process (the most widely used method today) or using programmable chips to produce cryptographic signatures. But as malicious actors continue to find new vectors of attack, these methods are proving insufficient.

Nigel Edwards, Hewlett Packard Enterprise security engineering fellow and vice president, argues that every subsystem on a computing device requires its own root of trust.

"Every active component—network interface cards, memory modules, storage controllers, power supplies—is running firmware or software by another name," says Edwards. "How do you know they're not infected with malware? Each of these needs to have a hardware root of trust."

The Distributed Management Task Force (DMTF), a technology standards organization, has been developing these types of protocols, which manufacturers can implement to allow subsystems to authenticate themselves. The group's Security Protocol and Data Model (SPDM) specification is relatively new, but products that utilize it are already being developed. The specification offers a roadmap for different components to measure and communicate with one another. (HPE is a board member of DMTF.)

"If we can ensure that all our NICs, memory modules, and storage controllers have SPDM, that makes these devices much harder to attack," says Edwards. "We're making the attack surface much smaller."

What threats do roots of trust help prevent?

One reason to implement roots of trust is to protect against counterfeit hardware—inexpensive clones of name-brand products, like phony Cisco switches. Fake components cost the electronics industry an estimated $250 billion annually in lost revenue, which doesn't include support costs incurred when the fake product fails and the real manufacturer is asked to fix or replace it.

More insidiously, adversaries may insert a hidden backdoor into a legitimate component that could allow them to steal data, spread malware, or hijack the device for use in future attacks.

Please read: Application and data security start in the supply chain

Roots of trust can also help protect poorly secured internet of things devices. In October 2016, the Mirai botnet was used to hijack more than 100,000 IoT devices and launch a massive distributed denial-of-service attack against domain name service provider Dyn, taking several major websites offline for most of a day.

If these IoT devices had a hardware root of trust enabled, the Mirai software's attempts to access and control them would have failed, says Tuyls, because untrusted code could not have been run on those devices.

When will hardware roots of trust be commonplace?

Tuyls says more than 300 million semiconductors have implemented Intrinsic's PUF technology, but that's still just a fraction of the billions of chips currently in use.

The first devices based on the SPDM specification are likely to start shipping before the end of 2022, says HPE's Edwards. But widespread use of hardware roots of trust in components is still years away, he warns.

"I would like to insist that every active component in every server we ship has SPDM and a hardware root of trust," Edwards says. "The reality is that we can't enforce that today, because not every component available to us has that capability. But over time they will."

Of course, as history has shown, when the computer industry finds new methods for hardening its defenses, attackers eventually uncover more sophisticated ways to gain access. Ben-Atar argues that managing hardware-related security risks needs to extend beyond verifying the authenticity of components to validating the entire manufacturing process, from design to distribution.

Even then, there's no silver bullet for cybersecurity, he notes. As defenses improve, attackers eventually learn how to evade them. But implementing hardware roots of trust might persuade them to look for an easier target.

"The worst thing that can happen is if you or your security team develop a false sense of confidence," he adds. "When your team tells you that you're fully secure, that's when you need to start worrying."