Functional Programming for Embedded Systems

Licentiate thesis, 2022

Embedded Systems application development has traditionally been carried out in low-level machine-oriented programming languages like C or Assembler that can result in unsafe, error-prone and difficult-to-maintain code. Functional programming with features such as higher-order functions, algebraic data types, polymorphism, strong static typing and automatic memory management appears to be an ideal candidate to address the issues with low-level languages plaguing embedded systems.

However, embedded systems usually run on heavily memory-constrained devices with memory in the order of hundreds of kilobytes and applications running on such devices embody the general characteristics of being (i) I/O- bound, (ii) concurrent and (iii) timing-aware. Popular functional language compilers and runtimes either do not fare well with such scarce memory resources or do not provide high-level abstractions that address all the three listed characteristics.

This work attempts to address this gap by investigating and proposing high-level abstractions specialised for I/O-bound, concurrent and timing-aware embedded-systems programs. We implement the proposed abstractions on eagerly-evaluated, statically-typed functional languages running natively on microcontrollers. Our contributions are divided into two parts -

Part 1 presents a functional reactive programming language - Hailstorm - that tracks side effects like I/O in its type system using a feature called resource types. Hailstorm’s programming model is illustrated on the GRiSP microcontroller board.

Part 2 comprises two papers that describe the design and implementation of Synchron, a runtime API that provides a uniform message-passing framework for the handling of software messages as well as hardware interrupts. Additionally, the Synchron API supports a novel timing operator to capture the notion of time, common in embedded applications. The Synchron API is implemented as a virtual machine - SynchronVM - that is run on the NRF52 and STM32 microcontroller boards. We present programming examples that illustrate the concurrency, I/O and timing capabilities of the VM and provide various benchmarks on the response time, memory and power usage of SynchronVM.