Intelligent Power Management in the World of Vision Systems

Capturing images has become an integral part of many everyday and not-so-common applications, from vehicle cameras through manufacturing inspection, to drones adding AI for machine learning and facial and gesture recognition. According to analysts MarketsandMarkets, the demand for video processing electronics is estimated to expand at a CAGR of 18.7% to around $3.2B by 2024, with security and surveillance expected to have the highest growth rates.

The sensors in video systems themselves, whether CCD or CMOS, have become miniature commodity items and can be low cost and low power, particularly CMOS types. Processing the video output, however, is often a complex operation and is increasingly done in-camera or at the point of use. For example, a wireless machine vision system might utilize “edge computing” to process data locally, providing faster response to actuators, less burden on centralized computing, and lower local power consumption with fewer wireless transmit cycles. Similarly, in the rapidly expanding surveillance drone market, local video signal processing is necessary to provide direct feedback to the navigation system for terrain analysis and to avoid collisions. In this particular example, small size, low weight, and minimal power consumption are vital for maximizing available flight time from the onboard battery.

 

Vision Processor ICs Need Multiple Power Rails

For the simplest applications, a vision processor, often a ‘System on Chip,’ might only consume around 0.5 W but still requires multiple voltage rails with their own power up/down sequencing requirements. These rails typically range from 3.3 V to sub-1 VDC for core, I/O, analog, memory, housekeeping, and connectivity functions. Figure 1 shows the rails for an example vision processor IC. Fully featured vision processor chips with 8 K quality video, complex, high-speed functions, and features such as multiple cores, extended interface options, debug facilities, a display controller, and ASIL security for automotive applications will consume higher power and require a higher number of rails. These applications require onboard power management techniques such as dynamic voltage and clock frequency scaling and multiple idle, “sleep,” and “deep sleep” modes to maximize battery life and flight time.

 

Figure 1. Typical vision processor “System on Chip” power rails. Image used courtesy of Bodo’s Power Systems [PDF]

 

The required voltage rails are powered from a local single-cell lithium-ion battery, and power conversion must be implemented efficiently for maximum battery run-time. The best solution to accomplish these goals is a Power Management IC (PMIC) which integrates multiple programmable and configurable DC-DC converters with remote configuration and control to achieve initial set-up, dynamic voltage scaling, and sequencing. A typical example of a fully featured PMIC is Qorvo’s ACT88760. The ACT88760 is ideally suited for video processing applications, but it can also power a wide range of processors, FPGAs, wearables, peripherals, and microcontrollers (Figure 2).

 

Figure 2. A Power Management IC suitable for video processor ICs. Image used courtesy of Bodo’s Power Systems [PDF]

 

The ACT88760 is designed for a 2.6 to 5.8V input, which matches single Li-Ion or Lithium-Polymer cell voltages, and features seven buck converters with current ratings from 2 A to 4 A. Pairs of bucks can operate in dual-phase mode for 8 A output, and each buck can be programmed from 0.5 V to 3.8 V. Six low drop-out (LDO) linear regulators are included, two of which can be configured as low-resistance load switches. In addition, ten general-purpose I/O (GPIO) pins are configurable for a variety of system functions which can also be accessed through an I2C interface. Configuration includes changing the output voltages, start-up times, output slew rates, system level sequencing, switching frequency, sleep modes, and other operating modes. The ACT88760 is an integrated converter and only requires a few external components. Qorvo designed the ACT88760 with a 2.25 MHz switching frequency to maximize load transient response and minimize the external component sizes. In deep sleep mode, the PMIC quiescent current is only 65 µA with a single LDO regulator enabled to keep the system “alive” but ready to respond to a “wake-up call.” Quiescent current is only 10 µA with all regulators disabled. The ACT88760 is available in a tiny 81-pin 3.85 x 3.85 mm WLCSP package to fit into the most space-constrained applications. The part is supported by the Qorvo ActiveCiPS dongle, which, in conjunction with the device evaluation board, allows user monitoring and configuration through an intuitive GUI without any special firmware or software. Having chosen default parameters for a particular end-product, the user can upload the configuration to Qorvo, who then ships preprogrammed parts with the required functionality.

 

Figure 3. The Qorvo “ActiveCiPS™” dongle. Image used courtesy of Bodo’s Power Systems [PDF]

 

PMICs for Battery Charge Control

For some battery-powered video applications, battery charge control is required, which can also be implemented in PMICs such as the Qorvo ACT81460. This device is suitable for lower-power home security camera applications and features two 0.4 A buck regulators, a buck-boost regulator, a boost regulator, and three LDOs, each rated at 100 mA. All outputs have programmable voltages and start-up/shut-down characteristics. The PMIC includes a 0.8 A linear battery charger with comprehensive charge control modes: trickle, pre-conditioning and fast charge, depending on the battery charge state. Battery loading can be as low as 2.1 µA with switching regulators disabled but still with internal monitors and references enabled. It provides four GPIOs and an I2C interface for monitoring and control, including dynamic voltage scaling. The device operates from 4 V to 5.5 V input while the battery range is 2.7 V to 4.5 V. As with the ACT88760, outputs can be sequenced and enabled via an I2C interface and configured using the using the ActiveCiPS dongle. Low power modes are implemented, including “sleep” and “deep sleep,” and the quiescent current is just 6 µA, even with six regulators and three load switches enabled.

PMICs are ideal companions for vision processors and SoMs where multiple tightly-regulated voltage rails are required, with sequencing and active control to minimize losses, both in the end-load and in the PMIC itself. As a bonus, standard PMICs that can be pre-configured and remotely controlled, such as those from Qorvo, can be used in multiple applications with consequent savings in purchased volumes and stocking costs.

 

This article originally appeared in Bodo’s Power Systems [PDF] magazine.