Reliable systems for micro aerial vehicles — SoC control solution

Editor's Note: Embedded designers must contend with a host of challenges in creating systems for harsh environments. Harsh environments present unique characteristics not only in terms of temperature extremes but also in areas including availability, security, very limited power budget, and more. In Rugged Embedded Systems, the authors present a series of papers by experts in each of the areas that can present unusually demanding requirements. A separate excerpt of the book addresses fundamental concerns in reliability and system resiliency.

Elsevier is offering this and other engineering books at a 30% discount. To use this discount, click here and use code ENGIN317 during checkout.

Adapted from Rugged Embedded Systems, Computing in Harsh Environments, by Augusto Vega. Pradip Bose, Alper Buyuktosunoglu.

CHAPTER 7. Reliable electrical systems for micro aerial vehicles and insect-scale robots: Challenges and progress (Cont.)
By X. Zhang, Washington University, St. Louis, MO, United States

4 BrainSoC

As described in the previous section, building the RoboBee system is truly a multidisciplinary undertaking, in this section we will zoom into its electronic subsystem and in particular dive into the custom designed system-on-chip (SoC) acting as the “brain” for RoboBee, which we nicknamed BrainSoC.


The ultimate goal of autonomous flight requires converting the external bench-top test equipment into customized electronic components that the robot can carry within its tight payload budget. Towards this end, we designed an energy-efficient BrainSoC to process sensor data and send wing flapping control signals to a power electronics unit (PEU) that generate 200–300V sinusoids for driving a pair of piezoelectric actuators to flap each individual wing [10]. Fig. 1 is a cartoon illustration of the connections between the BrainSoC and the power electronics.

To understand the basic operation of the power electronics, we first have to revisit the mechanism of exciting the flapping wings with piezoelectric actuators (Fig. 2). Electrically, these layered actuators can be modeled as capacitors. To generate sufficient wing flapping amplitude, these actuators need to be driven by a sinusoidal waveform of 200–300 V at the mechanical resonant frequency of the robot, approximately at 100 Hz. In one of our schemes, we fix the bias across the top and bottom layers of the actuator and use the PEU to drive the middle node of the actuator with a sinusoidal signal.

click for larger image

FIG. 1 Illustration of different functional components connected through the substrate flexible PCB, including the BrainSoC chip and the high voltage IC chip for the power electronics unit (PEU), in the RoboBee electrical system. (inset) The relative foot print of the components compared to a U.S. five-cent coin.

click for larger image

FIG. 2 (A) The voltage-driven piezoelectric bimorph that generates bidirectional wing-flapping forces in RoboBee; (B) the layered structure in the piezoelectric bimorph; and (C) the equivalent electrical circuit model for the piezoelectric actuator.

To provide the required drive signal in an efficient manner, a two-stage design is adopted for the PEU. The first stage is a tapped-inductor boost converter built with discrete components, and it outputs the high voltage bias VDDH in the 200–300 V range for the second stage. The second stage is implemented as a high voltage integrated circuit chip using double diffused metal oxide semiconductor (DMOS) transistors in a 0.8 μm 300 V BCD (Bipolar-CMOS-DMOS) process. It consists of two linear driver channels, connecting to the middle node of the left and right actuators, respectively. To generate the desired sinusoidal waveform for the drive signal, the controller of the linear drivers uses pulse frequency modulation (PFM) to encode the slope of the sinusoidal drive signals. Recall from the previous section that changing the shapes of these drive signals with respect to each other could result in roll, pitch, and, yaw rotation of the robot—shifting the drive signals up causes the robot to pitch forward, while applying different amplitudes to the left and right actuator results in roll; skewing the upward and downward slew of the drive signals result in yaw rotation (Fig. 3). As indicated by the block diagram in Fig. 4, the desired drive signal is generated by a feedback loop that relies on an embedded controller to calculate the exact sequence of the PFM pulses based on the current signal level and the high-level rotation command. More technical details on the design consideration and performance of the power electronic unit can be found in our recent paper [11].

We approach the RoboBee electronic system design with a multichip strategy, because the power electronics require specialized DMOS process that can tolerate extremely high breakdown voltages, while the real-time computational demand for autonomous flight control calls for faster digital logic circuits using more advanced CMOS process at smaller technology node. At the same time, since the multichip system has to fit within stringent weight, size, and power budget, we would like to minimize the use of external discrete components and to directly power off battery.

click for larger image

FIG. 3 Drive signals for the piezoelectric actuator to generate three degree-of-freedom maneuver— roll, pitch, and yaw.

click for larger image

FIG. 4 Block diagram of the two-stage PEU used for the RoboBee. (inset) The drive signal generated by the PEU for the piezoelectric actuator and the corresponding pulse modulation signals.

All these design constraints lead to the decision of a custom-designed SoC that embeds a multitude of functionalities into a single chip including integrated voltage regulator (IVR), analog to digital converters (ADC), multiple clock generators, and a computational core with heterogeneous architecture.


As the second chip in the multichip electronic system design for RoboBee, the BrainSoC (Fig. 5) integrates peripheral support circuits to obviate external components other than a battery.

First, RoboBee uses a 3.7 V lithium-ion battery as its power source, while the central control digital logics on the BrainSoC requires a digital supply voltage below 0.9 V. Due to the limited weight budget allocated for the SoC, neither external regulator module nor discrete components such as capacitors and inductors can be afforded, which leads to the integration of a two stage 4:1 on-chip switched-capacitor regulator as part of the SoC. The switched-capacitor topology is selected for the IVR because it does not need external inductors unlike the buck topology and it delivers much better conversion efficiency compared to a linear regulator, which suffers intrinsic low efficiency when the output voltage is only a fraction of the input voltage. The IVR steps the 3.7 V battery voltage down to 1.8 and 0.9 V, resulting in three distinct voltage domains.

click for larger image

FIG. 5 Block diagram of the BrainSoC used for the RoboBee.

Second, in addition to voltage conversion and regulation, clock generation is yet another function that must be entirely integrated for this microrobotic application. Our design budget does not allow room for adding external crystal oscillator component, and therefore all the clock signals have to be generated on chip. There are several different types of clock generators in the BrainSoC to fulfill different needs for the system: a high-precision 10 MHz relaxation oscillator sets the wing-flapping frequency to best match its mechanical resonance frequency; a 16-phase supply-invariant differential ring oscillator is designed for the switching control logics in the IVR that burns extremely low power and is directly supplied by the battery; multiple digitally controlled oscillators (DCO) generate the clock signals for the general-purpose microprocessor, memory, accelerators, and peripherals.

Next, the BrainSoC is equipped with a diverse set of peripheral circuits to interface with a multitude of sensor chips. For sensors with analog I/O channels, an embedded ADC is integrated that comprises of four 8-bit successive approximation ADCs channels multiplexed between 16 analog inputs; and for sensors with digital serial communication interfaces, we incorporate standard serial protocol controllers such as I2C, SPI, and GPIO.

Finally, the computational core inside BrainSoC employs a heterogeneous architecture. It consists of a 32-bit ARM Cortex-M0 microcontroller that handles general computing needs and acts as the master of the AMBA high-speed bus connecting to various memories. In order to meet the real-time performance demands of autonomous flight in an energy efficient way, we have to supplement the general-purpose computing system exploratory experiment, vision sensor and IMU are the essential sensing mechanism for RoboBee, accelerators for image processing and rotation control are built into the hardware with dedicated memory and coordinated by M0 via memory-mapped register.

The next installment from this chapter describes key issues for supply voltage management in a MAV SoC.

Reprinted with permission from Elsevier/Morgan Kaufmann, Copyright © 2016

Professor Zhang joined the faculty at Washington University in St. Louis in 2015. Previously, she was a postdoctoral fellow in computer science at Harvard University, where she worked on the RoboBee BrainSoC and energy-efficient computing projects. She has worked as a graduate research assistant at Cornell University studying variability-tolerant circuits. Zhang earned a doctorate in electrical and computer engineering at Cornell University in 2012. She earned a bachelor’s degree in electrical engineering at Tsinghua University in Beijing in 2006.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.