uWatt Sensor Sampling Solutions for Deeply Embedded Systems
To effectively utilize the limited power available in deeply embedded systems such as remote Internet of Things (IoT) nodes, exercising the most efficient design techniques possible is an absolute must. In this context, as challenging as it may seem, a real-world uWatt sensor sampling solution can be implemented today using an off-the-shelf MCU and a readily available low-cost sensor. To approach the design of such an incredibly low-power system, it is imperative to first establish the absolute lowest power standby mode of operation – as the normal mode of operation – and only then carefully add system features one by one with a smart approach of hardware and firmware.
To effectively utilize the limited power available in deeply embedded systems such as remote Internet of Things (IoT) nodes, exercising the most efficient design techniques possible is an absolute must. In this context, as challenging as it may seem, a real-world uWatt sensor sampling solution can be implemented today using an off-the-shelf MCU and a readily available low-cost sensor. To approach the design of such an incredibly low-power system, it is imperative to first establish the absolute lowest power standby mode of operation – as the normal mode of operation – and only then carefully add system features one by one with a smart approach of hardware and firmware.
The four most important steps to achieve a uWatt sensor sampling solution include the following:
- Establish the lowest power-sleep mode as the normal mode
- Focus on a minimalist active duty cycle
- Power-manage external sensors
- Energy aware firmware always
For discussion purposes a simplified remote temperature monitoring embedded system is shown in figure 1. An MSP430F2012 ultra-low power MCU is the embedded processor and a common negative temperature coefficient (NTC) sensor is used.
Figure 1 – remote temperature monitoring system
For analysis purposes the interesting current consumption parameters for the MCU are taken from the device datasheet and shown in table below.
| Parameter | 3V | 2.2V |
| Active Mode @ 1MHz | 300uA | 220uA |
| Sleep Mode @ 12kHz | 0.6uA | 0.5uA |
| Wake-up to 1MHz | <2us | <2us |
| ADC10 @ 200kSPS | 600uA | 520uA |
For the embedded system described, a first order total current consumption can be estimated by simply adding MCU Active Mode and ADC10 with the sensor-resister divider (150uA, 3V across two 10k ohm resistors) for a total of 1050uA, 3.150mW at 3V. This is system is definitely not an example of uWatt sensor sampling!
Importance of Seep-Mode
The active power for the MCU is attractive but by no means the best in the industry as many MCU vendors today will claim current consumption in the 100uA/MHz range and even lower. But what is more important is a useful sleep current of 0.6uA (at 3V) with a 12kHz oscillator, timer, BOR circuit, memory and all other peripherals circuits statically powered – this low sleep current combined with the ability to wake from sleep to a full active mode in under 2us, with all resources ready, means that performance is in essence, available on-demand. Either an active on-chip timer or any external pin can issue an interrupt and wake the device. No complex initialization or startup routine is required when exiting from sleep providing a high degree of system transparence to the designer by enabling the use of sleep mode as the normal mode of operation and active mode time slots selected as needed in extremely small bursts which can if needed be on order of micro seconds. In many cases the average active current approaches that of the sleep current of 0.6uA, 1.8uW at 3V.
Human Interface Duty-cycle
A typical sensor processing system that involves a human interface most often only needs to deliver a few hertz of bandwidth – people simply cannot respond to temperature or motion changes faster than this. A remote temperature sensor that is sampled at 5Hz is more than adequate. In contrast there is no logical reason why the MCU ADC10 has to convert at the maximum rated speed of 200kSPS when only 5Hz is needed. Turn the ADC10 off normally and only turn it on for a conversion when needed.
Stopping Current Leaks
With the circuit in figure 1 and the example discussed the sensor needs only to be powered when measured. If directly powered from the 3V source the sensor-resistor divider would consume 150uA permanently. For the example discussed, the divider only needs to be powered and measured 5 times/second and for narrow time intervals of approximately 10us. This equates to a needed duty cycle of only 50us/second or 0.005%. The easiest method to gate the power to this relatively high resistance divider, and stop a permanent current leak, is with any available MCU GPIO pin. The GPIO is set high to power the sensor divider only when needed, and low normally to de-power and completely eliminate all power leaks for this circuit.
The total power during the sensor measurement must also include that of the MCU’s ADC10 which is 600uA – operating continuously at 200kSPS. Thus the current of the ADC10 combined with the divider is a total of 750uA for the measurement function. When 750uA is multiplied by the 0.005% duty cycle the average contributed current consumption for the sensor sample function at 5Hz is reduced to only 0.0375uA, 0.1125uW at 3V. What is also important is that the ADC10 is designed for ultra-low power applications and does not require any measurable start-up time, it is available on demand.
Firmware
In addition to measuring the sensor, an amount of MCU active time is required to post process the sensor sample and is estimated at 100us. At 5Hz this equates to 500us/second or 0.05%. Assuming the active current of the MCU is 300uA, the contributed current consumption for post processing is only 0.150uA, 0.450uW at 3V.
From experience, the most significant impact on overall embedded system power reduction comes in the form of well thought out “energy aware” firmware. The software engineer must embrace the idea that every line of code that executes that does not have to, is a portion of the energy source that is eliminated that will never return.
Conclusion
For the sensor system shown, applying the four techniques described reduces the total consumption from 3.150mW to only 2.375uW at 3V – that is milli-watts to micro-watts. This is a power reduction of over 1000x. Nothing exotic is required, just common sense, an embedded processor specifically designed for ultra-low power and the judicious use power aware firmware. Operating the system at a lower 2V will further lower power, but such a voltage may not be available practically in some systems. In addition to the actual sensor sampling, an appropriate application specific output such as a low-power radio, LED or LCD must be added to the power budget.
