Energy consumption in modern microcontroller systems, part three: The impact of operating temperature on energy consumption

December 15, 2017 Horst Diewald and Uwe Mengelkamp, ProJoule GmbH

Energy consumption comparison of microcontroller/system on chip (MCU/SoC) systems – is one benchmark enough or do we need a parametric benchmark?

An important factor in the selection, market positioning, and success of a product is the energy consumption of the entire system. The traditional approach to measuring this is to express efficiency in microampere (µA) or microwatts per megahertz (µW/MHz), but this isn’t sufficient enough anymore. Energy storage systems store neither µA nor µW but Joules, which simply means energy. Comparing the energy usage of MCU/SoC devices therefore has become the prime focus of users.

Is one benchmark enough to select an MCU, an MCU family, or an entire MCU manufacturer? Are publicly available technical documents enough? How easy is it to choose the right vendor for your application?

In part three we look at exactly how much operating temperature can impact energy consumption.

The impact of operating temperature on energy consumption

Using the datasheet for an SAM MCU will help explain this in more detail. The datasheet shows temperature information for the following conditions, which is displayed in Figures 10 and 11:

  • VDDIN = 3.3V (values right side of curve) and 1.8V (left side of curve)
  • ULPVERG LPEFF Enable
  • RTC running on external 32KHz crystal
  • PD0, PD1, PD2 in retention state
  • BOD33 is disabled
 
   
Figure 10. The data from 02/2015 is based on simulations, while data from 06/2016 comes from characterizations. 

Figure 11. Here, the maximum values are roughly 2 – 2.6 over typical values. (Source: Atmel-42402E-SAM L22G / L22J / L22N_Datasheet_Complete-07/2016, Page 1152)

Figure 12 provides a visual representation of the impact temperature has on energy consumption for a SAM MCU in "STANDBY Sleep-Mode".

Figure 12. The current in RTC-Mode might be small at 20 ⁰C, but is significant at 85 ⁰C. The consumed energy is higher by a factor of 14! (Source: Atmel-42385J-SAM L21_Datasheet_Complete-06/2016, Page 1188)

The same datasheet also accounts for the following operation conditions:

• VDDIN = 0V (values right side of the curve) and 1.8V (left side of the curve)

• VBAT = 3,3V or 1,8V

• RTC running on external 32 KHz crystal

• BOD33 is disabled

This data serves as the basis for Figure 13, which defines "BACK-UP Sleep-Mode" for the SAM MCU.

Figure 13. The maximum values are roughly 1.5 – 2.5 times over typical values. The current ratio is about a factor of five between 85 ⁰C and 25 ⁰C. The current values at 85 ⁰C come close to the values in Figure 14. A power block structure is shown in the same datasheet on page 196; PDBACKUP includes a 32 kHz-crystal and RTC counter. Page 38 shows VBAT can power the crystal and RTC. (Source: Atmel-42385J-SAM L21_Datasheet_Complete-06/2016, Page 1152)

Figure 14 provides a visual representation of the impact temperature has on energy consumption for a SAM MCU in "Back-Up Sleep-Mode".

Figure 14. The current in RTC-Mode is small at 20 ⁰C but is – again – substantial at 85⁰C. The energy demand is a factor of five higher! (Source: Atmel-42385J-SAM L21_Datasheet_Complete-06/2016, Page 1189)

The energy increase is comparable between temperature changes from 0 ⁰C to 50 ⁰C and 80 ⁰C to 85⁰C. But how do you estimate data about current or energy consumption if no temperature data is available, or no data is available for higher temperatures? 

Measuring against benchmarks

The ULPBench-CP contains two (or even three) phases: operating mode (code execution) and sleep mode (RTC, clock). In addition, there are losses for the transition between modes. To estimate the impact of operating temperature without vendor data on energy consumption or battery life time, we evaluate existing public data from the ULPBench-CP (Figure 15).

Figure 15. Public data from the ULPBench-CP shows that temperature impacts the energy consumption of MCUs differently. (Source: Design und Elektronik)

The value of EEMark-CP benchmarks change a lot with temperature. The data in the table also indicates that the values in Figures 12 and 14 cannot be transferred to other MCUs/SoCs – the factor between 25 ⁰C and 85 ⁰C ranges from 1.16 to 10.73. Vendor data is needed to properly evaluate the impact of temperature on operating conditions.

Based on data from Figure 14 (SAML21, Rev. B) and Figure 15 (EEMark-CP: SAML21 Rev.B, LPEff on – 137,33) we can verify how much the EEMark benchmark value changes (Figure 16). From calculations, we take an operating energy of 3,72 µJ and sleep mode energy of 3,69 µJ. We assume the sleep mode current (RTC) is dependent on temperature, as shown in Figure 14. We further assume the energy demand in operating mode is independent of temperature.

Figure 16. Comparison of measured and calculated values for energy consumption confirms the assumption that the effects of temperature are independent of the operating current. 

This data allows us to conclude that energy demand increases if you apply a real-life temperature profile.

Temperature profile, a user example

A very basic review of a radiator can provide us with a good user example. For simplicity's sake, we assume three heating situations in one year, two transition phases (spring, fall), and a consistent heating period in winter. 

The heating profile is shown over 24 hours and is applicable for 90 days (one quarter). During the transition phases (Spring/Fall), we assume a loop of a heat up phase, a heating phase, and a heating pause overnight. During summer, no heating is assumed and the MCU operating temperature is around 25 ⁰C. In winter, we assume constant heating, but different temperatures at certain times (Figure 17).

Figure 17. The three simplified temperature profiles show the heating cycle of an example radiator application.

This provides a good baseline for evaluating how various temperature profiles impact the runtime or EEMark value of the SAM L21 MCU, since RTC values over various temperature ranges are available for the device.

The absolute energy consumption is going to be different based on the technical specification of the heat cost allocator. But you can calculate the incremental consumption of the MCU. It is roughly 32 percent higher than the comparative value providing a constant temperature of 25 ⁰C. The life time reduction can be estimated when other consumers of energy in the application are included to the calculations. 

Figure 19. Energy consumption according to different temperature profiles is illustrated using a SAM L21 MCU as an example. The energy consumption is shown as an average for the entire year by each hour in a day.

These calculations show that various temperature temperature profiles have a significant impact on the energy consumption of an MCU. The temperature dependency of the RTC mode and its affect on total energy consumption confirm this.

Conclusion

The effect of temperature on the energy consumption of an MCU can have a significant impact on the life expectancy of an application, particularly in systems with a limited energy source like a battery. This should also be considered when using an energy source with limited peak energy capabilities, like in energy harvesting systems or USB ports.

This article has demonstrated the need for suitable energy profiles. What you get for a Joule of energy is all that counts at the end of the day.

Horst Diewald is co-founder of ProJoule GmbH. At Texas Instruments, Horst invented the first ultra-low power 16-bit microcontroller, the MSP430. This product enabled battery applications to last more than 12 years from a single lithium ion battery. Horst owns and co-owns several patents around low-power architectures for MCUs/SoCs.

Uwe Mengelkamp managed the marketing and application organization for the ultra-low power 16-bit RISC MSP430 MCU at Texas Instruments. The charter was to extend MCU application life time with a given budget of energy. Later on Uwe developed the portable power market as business unit manager. Both roles focused on the maximum application throughput per joule of energy invested.

ProJoule GmbH

www.projoule.com

info@projoule.com

 

Previous Article
Using embedded FPGA as a reconfigurable accelerator
Using embedded FPGA as a reconfigurable accelerator

Embedded FPGA is a fairly new technology, but it is quickly finding a home in a range of applications. One ...

Next Article
MVTec brings deep learning to NVIDIA Pascal architecture

The deep learning inference in the new version 17.12 of the HALCON machine vision software was tested on NV...

×

Stay updated on processing and related topics with the Processing edition of our Embedded Daily newsletter

Subscribed! Look for 1st copy soon.
Error - something went wrong!