Verifying the Integrity of the Hardware

Supply Chain via Material Analysis

Mackenize Morris 

“This is a particularly pernicious threat … because it’s very difficult for the average citizen, company or government entity to understand every component that was put into a piece of equipment or network that they’ve purchased,” Homeland Security Secretary Kirstjen Nielsen. 

Abstract

Supply chain integrity is an exceptionally difficult problem to address when globalization has pushed manufacturing and shipping routes all around the world multiple times over. The ability to understand the life cycle of each component of a finished product has escaped the grasp of the consumer. The US Government already has recognized the dangers of securing its supply chain and put policies in place that prevent them from using hardware and software directly developed by countries on their sensitive country list; however, this is defeated by the difficulty of properly investigating third-party relationships within the direct vendor. Doing wholesale investigation further than one step down in the supply chain is almost an invasion of privacy, one which many vendors do not welcome. Without vertical integration and accountability, it is challenging to determine the legitimacy of both hardware and software components. These worries have been highlighted in the alleged SuperMicro hack when a rouge microprocessor was transplanted onto server motherboards during the fabrication process. Security experts believe there is no easy solution that easily safeguards against these threats; however material analysis can be repurposed by various methods of X-ray imaging to remedy the worry of supply chain interdictions.

Introduction

A series of National Defense Authorization Acts, as well as dozens of bills and provisions suggested in legislation, highlight the noted importance of securing both hardware and software supply chains for government organizations. Several departmental agencies, including the DOD, DHS, NIST, GSA, and DOE have raised rules pertaining to developing security controls revolving around supply chain. Without comprehensive investigations of supply chains and the inter-business connections between vendors, subcontractors, outsourced departments, and freelanced contracts it can be impossible to determine the origins of all hardware or software comments. The U.S. Government has seen supply chain security as an issue for their procurement process issuing guidelines on how to protect Information and Communications Technology (ICT) software. These controls fail to fully account the issue of physical tampering with computer hardware at the vendor level; something that grows more concerning with the degree of off source computer manufacturing. Supply chain interdiction is about introducing malicious components or firmware before the computer hardware reaches its customer. Believing governments and companies would protect their integrity in pursuit of commercial goals simply is not a reliable control for security, actions need to be taken to ensure that organizations can protect themselves regardless of whom interacts with their physical hardware. The easiest method of hardware tampering is to replace the firmware on one of the many microcontrollers that reside on the motherboard to take malicious action. Some organizations have a company policy to flash the firmware on new hardware components, which defeats this avenue of attack, but it remains an effective avenue of attack.

The alleged SuperMicro hack, which has been reported false by Apple, Amazon, and SuperMicro, the major companies named in a Bloomberg publication, all whom categorically deny that the hack ever took place, this still highlights the issue of, ‘what if a country is willing to use its power to influence a commercial enterprise to produce malicious hardware to facilitate its cyber-attacks?’  This method of hardware tampering would bypass all types of security and directly impact the functionality of the motherboard the microchip resided on; this attack defeats flashing the microcontroller firmware because the additional chip is hidden on the board to fulfill its malicious intentions. These attacks can only be carried out by the very large organizations and state actors whom have enough resources to influence the supply chain at its inception. CrowdStrike said in early 2018 report, “That 80% of cyber security personal interviewed believe supply chain to be of pressing concern.” Supply chain remains a vector for which little security exists, and all of that which does falls short of allowing customers to verify the hardware they received.

Figure 1: Computer manufacturing

Roughly 90 percent of the world’s personal computers are manufactured in China, as well as the majority of computer hardware giving their government an untold amount of potential influence over the companies that fabricate much of the hardware used in the world. Low manufacturing costs have led many computer hardware companies to off source their production facilities in East Asia or subcontract to facilities that reside there. Companies often subcontract or strike vendor agreements with organizations outside their direct oversight, this gap in visibility lends itself to vulnerabilities in the supply chain from hardware manufacturing to vendor processing.  A malicious chip can be introduced to the motherboard at the beginning of the supply line at which point it will, if disguised well, make its way to the customer. US officials report that as early as 2014 Chinese military has begun to orchestrate attacks via their hardware manufacturing industry.

Background

“You need specialized equipment and you have to carefully examine several heterogenous pieces of complex equipment. It sounds like a nightmare, and it's an expense that's hard for companies to justify.”

Supply chain hacks that involve introducing a small microprocessor to the motherboard with embedded memory seek to execute a list of commands to establish a remote command and control. Microprocessors that have been added to the motherboard to perform these malicious actions are incredibly effective because their ability to negate all software-based security controls. Unfortunately, due the relentless march forward by Moore’s Law and the advancement in technology, microprocessors the size of a grain of rice can carry more computing power than the rocket in the Apollo Mission. The saving grace is that microprocessors are all the same. They are fabricated in a laboratory setting to yield small scale transistors focusing on the largest amount of computing power for the smallest amount of space. This highly technical process cannot be easily replicated at home or effectively if deviations in production are made. Without jamming as many microscopic transistors onto the same chip the ability to conspicuously add it to a motherboard is lost. These hurdles to supply interdiction inform on possible methods of preventing attacks, by using the information at hand as a solution for testing system hardware components can be developed in order to determine whether a computer part is malicious.

The health of a system when regarding software is constantly measured by a comparison to a baseline. This involves comparing group policy, registry settings, number of users, software inventory, and various other software components of a computer network to a preexisting set of “defined normal”. These baselines are constantly the gold standard for the nominal state of a network, and with accurate knowledge of an enterprise’s network cyber security professionals can diagnose anomalous behaviors or potential compromise within the system. The same concept can be applied to the physical hardware that an enterprises system is comprised of, a sort of baseline for hardware. Using criteria developed from a known legitimate item, direct from the vendor, a golden image or baseline can be created. Against this testing criteria organizations can test their new hardware. This process suffers from the need of an image for each piece of hardware as well as attacks on the original gold image, which are identical to attacks on baseline images from vendors. However, this process does require a significantly larger number of “known goods” than its software equivalent. These are manageable, and after obtaining golden images, an organization will have all they need to verify their computer hardware against possible physical manipulation.

Material Science has yielded various material composition analysis techniques that can be refitted to solve the issue of maliciously interdicted components. Due to their fabrication process, microprocessors have a predictable silicon lattice architecture that repeats. This atomic structure possesses vastly different physical qualities than the surrounding printed circuit board which they reside on. While a possible microprocessor that could be embedded into a motherboard is small to the human perception, using tools that regularly measure material compositions on a basis of microns will not encounter major issues identifying the atomic structure of a microprocessor. Microprocessors consist mostly of silicon wafers with trace amounts of metals. The silicon atoms in the wafer bond to each other in a lattice structure consisting of Si-Si bonds that distinguish microprocessors from other personal components. This is a structure that is unique to microprocessors and microcontrollers. The background printed circuit board that backs up the backbone of motherboards consist mostly of Silcom Oxide (SiO2) with copper wiring for electric transference plus poly-epoxy, plastic compounds, for support. Several imaging techniques can distinguish between the microprocessor from other components on the motherboard. From these images a heat map can be produced allowing a comparison from component to gold standard. These techniques are not fool-proof, and they have limitations in speed, time, and cost; however, the technology exists that can be used for proof-of-concept work and demands the attention security practitioners.

Solutions

X-Ray Ptychography (XRP)

XRP is the holy grail because of material analysis; however, it has the highest barrier to entry being the most expensive solution to the issue at hand. It involves the need for a synchrotron, which is incredibly expensive on top of the electricity bill it takes to run that machine, but the benefit of such high-powered equipment yields 3-D resolution down to 14 nanometers. This technique is already used for reverse engineering Intel and AMD chipsets. The scope and range of the imaging technique would be expensive and take an unacceptablely long time to image an entire motherboard. It stands as a proof of concept or rather the placeholder of a distinct possibility for a perfect imaging technique. The development of this process is growing as industry leaders begin to gain insight into the benefits of gleaning inside of a competitors microprocessor. This technique took 24 hours to image a single Intel Chip, however with a more powerful X-Ray source or improved computing power for analysis that time could be reduced by 1000, making it feasible to scan entire computer hardware parts in under 30 minutes.

Figure 2: Diagram of Microprocessor

This image shows the level of detail that XRP can achieve. The individual logic gates on an Intel chip can be seen, as well as the 3-D layering of a chip. This imaging technique can look for embedded rouge hardware at a level of detail a human simply cannot. Figure2 details an almost nanometer by nanometer representation of an intel processor. With that level of detail applied to an entire motherboard, locating rouge hardware would be trivial. Researchers that developed this analysis for reverse engineering have suggested that it could be repurposed with the proper amount of effort for verifying against hardware trojans.

While the most expensive, time consuming technique, XRP gives insight into the power of material analysis; yielding 3D computer images of an Intel Chip on the sale of 100000 times smaller than a grain of rice. With repurposed future work XRP could yield results in computer hardware verification.

X-Ray Fluorescence (XRF)

XRF is a much better potential candidate because of its ability to identify different species on a larger and faster scale than other imaging techniques. X-Ray Fluorescence is a non-destructive method of elemental analysis of material in which a sample irradiated by an X-ray beam gives rise to some characteristic X-Ray fluorescence spectrums to which species of the sample can be identified. Since the interdicted agent is likely to be a microprocessor with a predictable atomic architecture the XRF scan can look for the X-rays that come off the surface of the motherboard. The XRF detector can be fine-tuned to highlight a specific wavelength or species rather than identify all atomic components. By hyper focusing into the silicon typical of microprocessors, identification of rouge hardware components can be reduced to simple XRF scans. The ability to hide a microprocessor on a motherboard is significantly diminished by the unescapable nature of microchip construction. XRF is a material analysis method, especially coupled with specialized equipment, that will allow organization to vet their own hardware without relying on the vertical integration of security within product supply chains.

Figure 3: XRF Scan

XRF can be used in unison with programable logic to generate a heat map of specific hotspots. This will allow for the determination of whether a malicious chip has been added to the hardware with low false positive rates. Since the silicon will highlight itself among the background the computer will produce obvious locations for a security expert to review when alerts are raised.

Figure 4: Heat Map of a Piece of Ore

Figure 4 shows the XRF heat image of a rock. The scale of highlighted atomic composition shows the detail that can be examined from a sample that experiences high degrees of entropy. Motherboards have predictable designs and this imagining technique can be used to pinpoint the design differences on the motherboard. The individual hotspots of Silicon can be picked out from amount the surrounding material with a high degree of accuracy.

X-Ray Fluorescence is a beneficially quick process.  Encapsulated XRF machines, while expensive, are not outside the purview of federal agencies, government contractors or Fortune 500 companies.  Smaller XRF detectors in the form of hand held devices can be used for species analysis, although a lack of necessary detail means these devices would be unable to detect a malicious component as small as the SuperMicro hack.  There is a potential for the creation of refined handheld devices that are specifically focused on the identification of Silicon hotspots.  Such a development would drastically reduce the barrier of entry for small and individual enterprise, in addition to increasing portability for mobile operations.  

X-Ray Computed Tomography (XRT)

 A common application of X-Ray Computed Tomography is the everyday CT scan. Tomographic imaging uses focused X-rays from multiple orientations to measure decreases in intensity along linear paths. This method of multidimensional measurements yields a 3-D layer by layer image of the object. These images are viewed in the form of “slices” which are cross-intersections of the object from whichever orientation the viewer chooses. The layer that is viewed corresponds to a certain thickness within the image, which is useful for cutting through the obscurity of the printed circuit board.

XRT is a relatively cheap and easy to deploy technology. Coupled with machine learning slice images can be analyzed for defects which differ from a baseline provided by the vendor. The planer images are high definition and does not suffer from and surface obfuscation an attacker could impose on hidden hardware. XRT technology has been used for quality analysis in the past and can be repurposed for hardware validation, while not specifically identifying silicon the outline of the hardware can be identified and run for additional hardware.

Figure 5: XRT slices of a PCB

These images show the power of the XRT image and the resolution they give in identifying hardware interdictions. Easily deployable and ready to use with almost no technical experience required, XRT has the potential to yield favorable results in the field of hardware integrity validation.

Conclusions

The possibility for supply chain interdiction increases as nation states face off against one another in a game of cat and mouse. To protect commercial and national interests as well as personal data, work needs to be done to secure the supply chain of computer hardware entering service.  Material analysis has the potential to examine hardware at a level that goes beyond a human’s ability. XRP, XRF, and XRT show that microprocessors have uniquely identifiable characteristics for material analysis   which can be leveraged for hardware verification. Using this technique in unison with gold image baselines will secure the supply chain from manufacturing to buyers. As long as the gold image stays intact, any level of supply chain interdiction will be detected by the user level analysis.

XRP, while one of the most detailed analysis techniques, is overly expensive with high barriers to entry that make it impractical for this repurposing of imaging techniques.  It shows the value of a concept, but costs make this an unobtainable goal expect in the rarest of circumstances.

XRF technology has the potential to be ported to handheld devices which will allow any size organization to scour their hardware for rouge microprocessors only needing a heat map picture of a legitimate hardware comparison. Currently XRF   can be repurposed for this type of imagine analysis, although the cost is still high, yet not unobtainable for large organizations who value their data.  

XRT is currently the cheapest and most readily available technology to be deployed for the purposes of hardware validation. With ample work the process of validating hardware via programs that detect defects in the PCB.