How to detect hardware implants on circuit boards

A hardware implant is an extra component, or a quietly altered one, placed on a circuit board so a device does something its owner never asked for. Finding one comes down to comparing what is on the board, and inside it, with what should be there. This guide covers how researchers detect spy chips and other PCB tampering, what each method misses, and what an organisation can check before hardware goes into service.

What is a hardware implant?

Researchers usually call malicious hardware changes "hardware Trojans". A 2020 University of Florida paper on board-level Trojan benchmarking describes two types: an Addition model that "requires the addition of new components", and an Alteration model that "alters the existing PCB layout without adding new components", for example by changing traces. Triggers are designed to "minimize the probability that the Trojan gets accidentally triggered during functional testing".

Board-level implants are not exotic. At the 35C3 conference in 2018, Trammell Hudson presented a proof-of-concept implant that sat between a server's baseboard management controller (BMC) and its flash chip and altered data as it was read. In 2019, Monta Elkins altered a Cisco firewall with "a $150 hot-air soldering tool, a $40 microscope, and some $2 chips ordered online", according to Wired's report as quoted by Bruce Schneier.

The "Big Hack" report: what was claimed and what was disputed

On 4 October 2018, Bloomberg Businessweek reported that tiny spy chips had been added to Supermicro server motherboards, affecting almost 30 US companies including Apple and Amazon, Engadget summarised. According to The Register, Bloomberg alleged that "Beijing's military intelligence pressured or bribed a Chinese manufacturing subcontractor" of Supermicro.

The companies denied it. Apple said it "has never found malicious chips, 'hardware manipulations' or vulnerabilities purposely planted in any server". Amazon Web Services said that "at no time, past or present, have we ever found any issues relating to modified hardware or malicious chips". The US Department of Homeland Security and the UK National Cyber Security Centre both said they had no reason to doubt the companies, as The Register reported.

Bloomberg stood by its story. In February 2021 it published a follow-up saying the original "only captured part of a larger chain of events", and Supermicro called the new piece "a mishmash of disparate and inaccurate allegations that date back many years", Fortune reported. Treat the report as disputed, not proven.

Detection methods and their limits

Optical inspection against the design

The most direct check compares the physical board with its design: the bill of materials (BOM), which lists every part, and the layout, which shows where each part and trace belongs. After the Bloomberg report, the University of Florida's FICS Research lab described a system that compares a board and its components against the intended design using optical scans, microscopy, X-ray tomography and artificial intelligence. Its director, Mark Tehranipoor, was quoted as saying it "could have identified this [rogue] part in a matter of seconds to minutes". A 2020 review of the Big Hack he co-authored stresses the need for new visual inspection techniques for PCB assurance.

Optical methods only see the surface. The 2020 board-level Trojan paper notes that when a reference design exists, "the attacker is likely to target alterations within the internal layers to remain stealthy". Hudson adds that spare pads left for board variants make it hard for an outside tester to tell intentional parts from suspicious ones.

Comparison with a golden board

A "golden" board is a unit known to be clean, used as the reference for images or measurements. A 2007 IEEE paper on Trojan detection using IC fingerprinting built chip fingerprints from power, temperature and electromagnetic measurements of a few invasively tested chips, then checked the rest of the batch statistically. Simulations suggested Trojans three to four orders of magnitude smaller than the main circuit could be detected.

The weak point is the reference itself. A 2024 preprint by Safa, Schaumont and Tajik says a physical golden sample "is not feasible in many real-world scenarios", partly because a trustworthy assembly factory would be needed to make it. They report evidence that a signature simulated from the PCB design files can stand in, provided the approximate parasitic impedance values of the components are known.

X-ray and CT inspection

X-ray imaging sees through the board. Hudson writes that "X-Ray inspection would detect any implants between layers as well as any active components that are masquerading as passives", though it would not necessarily separate counterfeit chips with identical markings.

The cost is time. A 2013 Lawrence Livermore National Laboratory report notes that "as with all CT systems, users must trade off spatial resolution, inspection volume and imaging time". For nanometre-scale imaging of a chip, it estimated about eight days to scan a 0.5 cm area of interest, even with planned speed-ups, and recommended scanning suspect sub-volumes instead of the whole chip.

That report lists X-ray CT among the techniques investigated under DARPA's TRUST in Integrated Circuits programme. Contractors were sent chips containing unspecified malicious circuitry, with a goal of a 90 percent detection rate, IEEE Spectrum reported in 2008.

Electrical and side-channel testing

Functional tests check that a board does what it should, not that it does nothing else. As Princeton's Ruby Lee told IEEE Spectrum, "You don't check for the infinite possible things that are not specified." Side-channel tests go further by measuring power, current or electromagnetic emissions and looking for differences.

Small differences are easy to miss. In the 2020 board-level study, the implants raised current consumption by at most 3.5%, and the authors concluded that without a reference signature "such minor increase in current consumption is unlikely to raise any suspicion". Newer work measures the impedance of a board's power delivery network instead, but it still needs a reference to compare against.

Firmware checks

Firmware is the code stored on a board's chips that runs beneath the operating system. NIST's Platform Firmware Resiliency Guidelines (SP 800-193) group defences into protecting firmware against unauthorised changes, detecting changes that occur, and recovering rapidly and securely. NIST's SP 1800-34 shows how organisations can check that laptop and server components "are genuine and have not been tampered with", using data vendors store in each device.

Firmware checks have a blind spot. Hudson notes that reading the flash chips with an external programmer won't reveal his implant, because its data is not stored in the flash. In his words, "the hardware can often lie". Firmware checks and physical inspection cover different ground, so use both.

Detection methods compared

What to check before you deploy hardware

In 2022, Australia's Information Security Manual (ISM) added controls ISM-1790, ISM-1791 and ISM-1792, asking buyers to verify integrity "as part of acceptance of products and services", iTnews reported. The ISM notes that "ICT equipment may benefit from tracking and tamper-evident packaging" and that "a counterfeit product or service securely delivered is still a counterfeit product or service".

1. Get the reference first. The ISM says suppliers "should be consulted on how best to confirm the integrity of their products and services". Ask for the BOM, layout files and a known-good firmware version.
2. Inspect every board, not a sample. IEEE Spectrum described technicians picking a single chip from a batch of thousands for inspection, assuming the rest are essentially identical.
3. Send anomalies to X-ray. Aim imaging at suspect areas instead of scanning everything.
4. Verify firmware against the known-good version before first boot on your network.
5. Seal and record. Use tamper-evident seals or packaging so later tampering shows.

How BoardSafe fits

BoardSafe uses the design-comparison approach. Its scanner checks each board against its original design files and flags parts that are fake, swapped or not in the design, such as an extra chip, then checks firmware against a known good version. It takes under 60 seconds per board, found 97% of planted faults in internal testing, and runs on site, so design files stay with you. Devices that pass get a serial-numbered tamper-evident seal across the join of the case. See how the scan works.

For the wider picture, read our guides to hardware supply chain security in Australia and firmware integrity verification.

Questions people ask