My Smartphone Knows What You Print

My Smartphone Knows
What You Print
Exploring Smartphone-based Side-channel Attacks Against 3D Printers
Chen Song, Feng Lin, Zongjie Ba, Kui Ren, Chi Zhou, Wenyao Xu
Presented By: Jack Barker
Overview
• Investigates using side channel methods to determine what is
being 3D printed.
• Uses a Nexus 5 smartphone and it’s sensors
• Is it possible to infer IP information when a smartphone is placed
nearby and record side-channel signals during the 3D printing
process?
3D Printing
• A form of additive manufacturing
• Objects are CADed
• CAM (computer aided
manufacturing) software is used to
slice the object into multiple layers
• Builds an object up layer by layer by
extruding hot plastic
• Three axis of movement, X Y and Z
3D Printing
• Used in industry widely for prototyping currently
• Fast and cost efficient with less waste than alternatives
• Printers are accessible and affordable, making them widely used
• By 2021, it is estimated that the market will be worth $20 billion
• Increase in use means increase in IP sensitive products being printed
• Need to consider security of printed objects
Side Channel Attacks
• An attack based on information based in information gained from
physical implementation
• Acoustic and Magnetic signals are considered
To gather IP from a 3D printer:
• X,Y,Z movement and extrusion must be gathered
Information Gathering
Acoustic side channel:
• Each axis of movement has its own distinguishable sound
• Each motor is in different structures
• Used to determine speed
Magnetic side channel:
• Patterns in magnetic fields are observed
• Magnetic data in each coordinate changes when motor activated
• Used to determine direction
Support Vector Machine
• Information gathered from acoustic and magnetic analysis is used
in an Support Vector Machine (SVM)
• Supervised machine learning model
• Needs to be trained for a specific printer
• Experiment trained with 1000 audio frames and 2000 magnetic
frames
• This is important in converting the data gathered back to G-Code.
Processing of Data
• Signal noise needs to be removed as well as white noise.
• A Savitzky-Golay (data smoothing) filter is used on collected data.
Layer Movement Analysis
• Determine which plane the movement is on (X/Y plane or vertical)
• Acoustic analysis used
• Z axis is not belt driven, distinct acoustic sound
Processing of Data
Header Movement Analysis
• Extruder feeds filament through at a constant speed.
• Extrusion speed determined by the layer height and material.
• When not extruding, printer moves faster
Acoustic used to determine if there is extrusion in a frame
Processing of Data
Axial Movement Analysis
• When on XY plane, need to determine which axis
• Acoustic channel is again used to determine which axis movement
is on.
Directional Movement Analysis
• Need to determine direction of movement
• Magnetic channel is used to determine this
Accuracy
• Authors devised a model, Mean Tendency Error (MTE) to determine
how accurate a reconstructed object is
• MTE assesses a geometric reconstruction of shape difference using
points
• Calculates the directional consistency between design and
reconstructed design
• A metric that takes points in the design and the reconstructed
design and takes into account the error of points
IP Reconstruction
• We must convert the gathered data to G-Code
• Using the information from the SVM, the authors could recreate
the following
IP Reconstruction
Issues
• Distance
• Print Speed
• Phone Position
• Ambient Noise
Defences
• Dynamic Path Planning
• change speed during print
• Dummy Task Injection
• Dummy movements outside object
• Hardware Shielding
• Shield sound and magnetic information from intruders
• Side Channel Interference
• Home appliances (eg, microwave) can produce magnetic field
Questions?