AI: Edge Computing

Practically Speaking: ML Application Use Cases

The goal of AI-based algorithms running on embedded systems is to process real-world data collected by sensors in ways that are more efficient than more allowed by the more traditional procedural or object-oriented programming methodologies. Perhaps the most visible use case in our collective consciousness is the progression from legacy automobiles, to cars with automation assistance–such as lane departure warning alerts and collision avoidance systems–to the ultimate goal of self-driving cars with no human in the control loop. However, many other less-conspicuous uses of deep learning are already being used even if you did not know it. Voice recognition in your smartphone or virtual assistants such as Amazon Alexa leverage deep-learning algorithms. Other uses include facial detection for security applications and or background replacement, sans green screen, in remote meeting software such as Zoom.

The Workflow

According to the documentation provided for TensorFlow Lite for Microcontrollers, the developer workflow can be broken down into five keys steps (Figure 2). These steps are:

1. Create or Obtain a TensorFlow Model: The model must be small enough to fit on your target device after conversion, and it can only use supported operations. If you want to use operations that are not currently supported, you can provide your custom implementation.
2. Convert the Model to a TensorFlow Lite FlatBuffer: You will convert your model into the standard TensorFlow Lite format using the TensorFlow Lite converter. You might wish to output a quantized model since these are smaller in size and more efficient to execute.
3. Convert the FlatBuffer to a C byte array: Models are kept in read-only program memory and provided in the form of a simple C file. Standard tools can be used to convert the FlatBuffer into a C array.
4. Integrate the TensorFlow Lite for Microcontrollers C++ Library: Write your microcontroller code to collect data, perform inference using the C++ library, and make use of the results.
5. Deploy to your Device: Build and deploy the program to your device.

Some caveats that a developer should be aware of when selecting a compatible embedded platform for use with TensorFlow Lite libraries include:

1. 32-bit architecture such as Arm Cortex-M processors and ESP32-based systems.
2. It can run on systems where memory size is measured in the tens of kilobytes.
3. TensorFlow Lite for Microcontrollers is written in C++ 11.
4. TensorFlow Lite for Microcontrollers is available as an Arduino library. The framework can also generate projects for other development environments such as Mbed.
5. No need for operating system support, dynamic memory allocation or any of the C/C++ standard libraries.

Next Steps

Google offers four pre-trained models as examples that can be used to run on embedded platforms. With a few slight modifications, they can be used on various development boards. Examples include:

• Hello World: Demonstrates the absolute basics of using TensorFlow Lite for Microcontrollers.
• Micro-Speech: Captures audio with a microphone to detect the words "yes" and "no".
• Person Detection: Captures camera data with an image sensor to detect the presence or absence of a person.
• Magic Wand: Captures accelerometer data to classify three different physical gestures.

Visit the additional articles below for a series of step-by-step guides that will show you how to get these models working on various different microcontroller platforms.

• Adding A.I. Magic
• Build a Voice Assistant with TensorFlow Lite
• AI on the Edge
• AI on the Edge

Securing a Device

To look at a device and understand how it can be exploited, we look at what’s called the attack surface. The attack surface for a device represents all of the points where an attacker can attempt to exploit or extract data from a device. This attack surface could include:

• The network ports that interface to the device
• The serial port
• The firmware update process used to upgrade the device
• The physical device itself

Attack Vectors

Communication

Attacking an interface or protocol is a multi-layered issue. There is the security of the communication itself with the cloud—including data security—as well as access security to the device through one or more protocols such as HTTP.

The Transport Layer Security (TLS) should protect all communication to and from the device. This type of cryptographic protocol covers authentication—to ensure that both sides can be specific about who they are communicating with—as well as encryption of all data to avoid eavesdropping attacks. This is ideal for an edge device that communicates to a remote cloud over public networks like the internet.

Given the speeds at which data moves over IP networks, hardware acceleration is a must in order to efficiently manage authentication and data encryption and decryption. Processors with hardware encryption acceleration like the TI EK-TM4C129EXL include on-chip crypto acceleration for TLS, ensuring secure communication with remote systems.

Using protocols like Kerberos for authentication can ensure that a client and server securely identify themselves. Kerberos relies on symmetric-key cryptography or public-key cryptography, both of which can be accelerated using processors that include cryptographic engines.

Protocol Ports

The protocol ports used with a network interface form one of the largest attack vectors on an internet-connected device. These ports expose protocol access to the device—for example, a web interface is exposed typically through port 80—and therefore provide information to the attacker on types of exploits to attempt.

One of the simplest ways of protecting these ports is with a firewall. A firewall is an application on a device that you can configure to limit access to ports in order to protect them. For example, a firewall can include a rule that prohibits access to a given port except for a predefined trusted host. This limits access to the port and helps to avoid common attacks using protocol exploits such as buffer overflows.

Firmware Updates

Edge devices are becoming increasingly complex, performing more advanced functions than prior generations—including machine-learning applications. With this complexity comes the requirement to fix issues and release updates to devices. But, the firmware update process creates an attack vector. By implementing security measures for firmware updates in your edge security plan, you can mitigate the risks posed by attackers.

Code signing is a common security method used to avoid malicious code from entering a device. This entails digitally signing the firmware image with a cryptographic hash, which can be used on the device prior to the firmware update process, to ensure the code is authentic and has not been altered since the signing process.

The signed code can also be used at boot-time to ensure that the firmware in the local storage device has not been altered. This covers two attack vectors, attempting to update the device with an exploited image using the device’s update process and protecting the device against an image forced into the local storage device.

The processor used within a device can be helpful here particularly if it implements a secure cryptographic engine for hash generation and checking. One example is the Microchip CEC1302 that includes a cryptographic Advanced Encryption Standard (AES) and hash engine.

The use of a Trusted Platform Module (TPM) is also beneficial. The TPM is a secure crypto-processor dedicated to security features and typically includes hash generation, key-storage, hash and encryption acceleration, and a variety of other features. One example is the Microchip AT97SC3205T, which implements a TPM in the context of an 8-bit microcontroller.

Physical Security Measures

Creating tamper-proof designs can help detect if the device has been physically opened or compromised in some way. This also includes minimizing external signals whenever possible to limit the ways that an attacker can monitor a device in their possession and identify exploits. Attackers may attempt to monitor bus signals to identify secure information and, in extreme cases, may apply temperature changes to the device, change clock signals, and even induce errors through the use of radiation. Understanding the methods that a motivated attacker will use to understand your device will help in building more secure products.

Where to Learn More

With the state of cyber warfare today and the plethora of motivations that individuals and states seek to exploit devices, edge security is an uphill battle. But, implementing modern security practices and considering security at the start of product development will go a long way to keep your device safe. Early analysis of the attack surface of a device helps to identify where to focus attention in order to create more secure devices. You can learn more at the Mouser Security blog.