This AI Startup Keeps Counterfeit Components Out of Your Gadgets

Share on linkedin
Share on facebook
Share on twitter

Everything in electronics is made of components, capacitors, microprocessors, resistors; chips essentially. If one component in a system fails, the device won’t function as intended. For a smartphone or connected IoT gadget, the adverse effect of a component failure caused by counterfeits isn’t of much of the public’s concern. But for high-grade electronics controlling high impact devices, like those found in MRI machines, ballistic missiles, low-orbit satellites, or commercial aircrafts, with virtually no margin for error, counterfeits can wreak havoc. 

The damages incurred by counterfeit components is estimated at $580 billion annually, including instances of programmable components laced with malicious malware – hardware cyber attacks, which can cost a mere $200 to execute. Today, the overwhelming majority of the components are assembled without any prior authentication testing, paving the way to a lucrative $30 billion fraudulent counterfeit components industry. 

Counterfeit components consist of cheap copies, recycled, rejected, old, or tampered components. They may hold up to the first functionality tests after assembly however their stark inferiority manifest with performance features. For example, cheap copied capacitors won’t maintain capacitance when operating under voltage bias, and old components will cause increased failures rates. There are inspection and screening tools available today to catch counterfeits, like a Scanning Acoustic Microscope to discover ‘black-topping’, or a Scanning Electron Microscope that analyzes metrology and abnormalities. But there’s no practical way to conduct functional tests for all the components on the circuit-board under all possible operating conditions. Israeli AI startup Cybord is tackling this endemic market with computer vision based algorithms to detect counterfeit components before they’re assembled.

Cybord is developing a patent pending software system that allows affordable scanning of 100% of a manufacturer’s components and detection of re-programmed hardware components, before the damage is done.

Cybord’s co-founders (left to right): Gil Givati and Eyal Weiss.

The startup was founded in late 2018 by Gil Givati and Eyal Weiss to automate mass component inspection and authentication at industrial scale. Givati, CEO, is a seasoned executive with almost two decades of technology development and management experience, specialized in image processing. Weiss, CTO, holds a PhD in electronics and computer engineering, with experience in nuclear research, plasma physics and AI software and hardware theorem. Weiss was also the recipient of the medal of honor last year by the Israeli military for his work on a classified sensing project.

“Cybord delivers affordable, fully automated, multi-physics, high throughput, 100% non-destructive component inspection and authentication” explained Givati. Their system, which integrates into the assembly SMT machine, scans every component without removing it from its original packaging and independently assures and reports that only authentic, fresh and untampered components are used on each assembled board. 

Surface mount technology (SMT) component placement systems, or pick-and-place robots are used to place surface-mount devices onto a printed circuit board. They’re equipped with a vacuum probe to manipulate the components and video cameras for top and bottom views of the components assembled. 

“We are now training our models with manufacturers in Israel. Our solution is based from a mix of 20 algorithms, using supervised and unsupervised learning computer vision techniques, including GANs and classifiers that identifies the counterfeit components as well as searches for evidence. Each sub-algorithm addresses different aspects of the classification problem, looking for anomalies in the chip, the soldering leads or classifying components by the leads’ age. We can identify where it was manufactured, if it was tempered with, all providing a holistic view of the components being used” explained Weiss.

Ribbon reels containing components are fed into the pick and place machine to be integrated onto the board.

Their unsupervised learning technique focuses on the ribbon reel of components as a whole, comprised of up to 10,000 components per reel. Each component manufacturer maintains unique parameter difference tolerances (i.e. radius of edges, pitch of soldering, full width half maximum) from competing components which Cybord estimates in order to verify its authenticity.

They built a ribbon reel scanner device in order to capture the same images as captured by the SMT machine, albeit before they’re mounted for production. “We scanned 2.5 million components to date, using a 5 megapixel color camera, providing our database for algorithm training” explained Weiss. Everyday, they scan an average of 50,000 components.

The startup is targeting electronic manufacturer services (EMSs) and Original Equipment Manufacturers (OEMs) in sectors like military, mobility, autonomous vehicles and aviation, largely dominated by EMS market leaders Flex, Jabil, and Samina. The startup licenses the use of their algorithms through an API to detect counterfeit components based on the model of the component. If a component is found to be counterfeit, it’s flagged and discarded.

They’re currently raising their seed funding of $2 million and recently graduated from Sigma Labs Accelerator. They received a grant from the Israel Innovation Authority, and are collaborating with the Israel Cyber Authority, to protect government infrastructure against cyber attacks. They’re also recently signed a new customer contract with a local manufacturer in Israel operating for Elbit Systems.

“The market is based on trust within the supply chain,” explained Givati. “But on average there’s at least one counterfeit component in any electronic device and most manufacturers use counterfeit components unknowingly. In fact, 5-10% of counterfeit components are identified as quality issues incorrectly. But they’re really counterfeit issues that failed statistically later on in the product life cycle and EMSs are seldom informed, dramatically tarnishing their customer trust. Even if a component is recalled, they’re rarely traced back to the couterfeit and are signed off as a business cost” said Givati. “With Cybord, this problem is tackled by analyzing every component and this is truly a great leap in cybersecurity and safety for components.” 

Continue Reading