Alif Processors Power Saving Features


The Ensemble family of power saving processors from Alif Semiconductor are designed to be energy efficient with an architecture capable of deployment in battery powered environments. Alif processors use energy efficient cores capable of waking up to perform tasks and sleeping to save energy. This contrasts with high powered devices that are always-on and must stay plugged in to a power source.

Alif’s multicore processors are equipped with high-frequency and high-efficiency cores separated by operating regions which then clock gate or power off. Industry leading battery life can be achieved by employing the power saving features such as running at high or low frequencies, clock gating as necessary, powering off unused regions, and sleeping for extended periods. This application note will detail the power management architecture of the device and explain how to utilize the features in the
design of your application.

The Ensemble power saving family scales from single-core microcontrollers to multi-core fusion processors. The fusion processors combine up to two Cortex-M55 microcontroller cores, two Cortex-A32 microprocessor cores, and two Arm Ethos-U55 microNPUs.

Power Saving Features Overview

Operating Regions

Alif Semiconductor Ensemble Fusion Processor Operating Regions
Figure 1 Alif Semiconductor Ensemble Fusion Processor Operating Regions

All processor components are encapsulated in one of three operating regions: High-Performance, High-Efficiency, and AON (always-on).

High-Performance Region: The highest clocked cores and memories will reside in the High-Performance Region since they run at higher frequencies ranging from 400MHz to 800MHz and typically require a PLL supplying the clock source. Within the High-Performance Region there are two processing subsystems of interest:

  • APSS: two Cortex-A32 microprocessor cores (up to 800 MHz)
  • RTSS-HP: Arm Cortex-M55 MCU and Ethos-U55 microNPU (up to 400 MHz)

Also within the High-Performance region is PD-6, the SYST power domain. This subsystem contains many of the central device resources shared between the processing subsystems. These shared resources are:

  • Two SRAM memory banks of up to 4 MB + 2.5 MB
  • Non-volatile memory: MRAM on-chip, OSPI and SDMMC off-chip
  • System peripherals (excluding LP- peripherals)

High Efficiency Region: This region is lower clocked than the High-Performance region and the cores can be used for simpler tasks and connectivity services while the High-Performance region is powered down. Developers can choose to perform many tasks in this region, using lower power and lower frequency components, without needing to wake the High-Performance Region. Processing subsystems of the High-Efficiency Region are as follows:

  • RTSS-HE: Arm Cortex-M55 MCU and Ethos-U55 microNPU (up to 160 MHz)

Battery Always On Region (AON): This contains the shared RTC, LPTIMER, LPGPIO, LPCMP, and 4KB of Backup Memory. The region is always running when a battery or sustained power source is connected to the VDD_BATT supply. The processor can be placed in an extreme power saving mode, called STOP Mode, where the High-Performance and High-Efficiency Region are powered off leaving only the AON Region.

The operating regions are summarized below in Table 1 Processor Operating Regions. Note that each processing subsystem may contain memories and peripherals within it, i.e., local resources, that are separately clocked and powered from the centrally held shared resources in PD-6 SYST. Please refer to the “Power Supply Management” section of the device datasheet for more information on the operating regions, subsystems, and power domains.

Battery Always-On Region
Lowest Frequency
Lowest Leakage
High-Efficiency Region
Lower Frequency
Lower Leakage
High-Performance Region
Higher Frequencies
Higher Leakage
Power Domains 0 and 1Power Domains 2, 3, 4Power Domains: 5-9
Table 1 Processor Operating Regions

Power Modes

Embedded systems are often interrupt driven, meaning incoming signals from the system are used to notify the CPU when data is ready or that a task should be started. Then, when the task is complete, the CPU enters an idle loop waiting for the next interrupt. In a battery powered system, developers avoid idle loops which consume unnecessary energy so processors will instead clock gate or power off the CPU in between tasks. In Alif processors, each processing subsystem’s CPU can do the same by operating in one of one of three CPU States: RUN, SLEEP, or OFF. Developers can choose between these states, described in Table 2 CPU States Summary, and use them to affect the power consumption of the device.

When a core in a processing subsystem is running and executing code it is in the RUN State. Any code execution, while loops or “no-op” instructions will keep the core’s clock running. Power consumption is highest in the RUN State. Developers can choose to stay in this state for continuous polling or for time critical applications where application response should be as quick as possible. In a power sensitive application, a core can reduce its power draw compared to the RUN State by entering the SLEEP State where its core clock will be stopped. During the SLEEP State, the CPU’s core clock will be stopped. Developers can choose this state to stop all code execution until an interrupt event wakes the core again. Since no power is lost in the subsystem, the core has retained its context and can resume the RUN State quickly.

Developers interested in more significant power savings will have a core and its processing subsystem further placed in the OFF State. The CPU core cannot power off independently of its local subsystem. Any resources in the subsystem are powered off along with the CPU core.

All processing subsystems can change between the CPU States listed from Table 2 CPU States Summary below and they can do so independently of each other. For example, one RTSS may be off while the other RTSS is awake capturing sensor data. The Secure Enclave Subsystem (SESS) maintains its own sleep and wake cycles. Its behavior is dictated by a combination of Alif System Software and any configuration input by user software. Generally, developers should aim to architect their applications to minimize time spent at the highest states and allow for power saving sleep cycles in between tasks.

CPU StateDescription
RUNWhen a CPU core’s clock is running, typically executing code.
SLEEPWhen a CPU core’s clock gated and can quickly resume.
OFFWhen a CPU core and its subsystem are powered down.
Table 2 CPU States Summary

As mentioned, the processing subsystems are meant to independently change CPU States without coordination with each other. Beyond these CPU States, Alif has defined additional SoC Power Modes which affect the central device resources shared between each processing subsystem. The overall SoC is in GO Mode as long as one processing subsystem is in the RUN or SLEEP State. In GO Mode all central device resources (clocks, memories, peripherals) are available for use by any of the processing subsystems. The device will stay in GO Mode for as long as any shared resource is being utilized. It is up to the SESS to determine when to change SoC Power Modes. When all processing subsystems are off and shared resources are not utilized then the SESS will be able to transition the SoC to STANDBY or STOP Mode.

Device Power ModeWake-Up Peripherals
GOOne or more Processing Subsystems are in RUN or SLEEP stateAll
READYAll processing subsystems are in SLEEP state.All
IDLEAll processing subsystems are in OFF state.All
STANDBYAll Processing Subsystems and shared resources in the High-Performance region are powered down.All GPIO, LPUART, LPI2C, plus STOP Mode wake-up peripherals
STOPAll Processing Subsystems and shared resources in the High-Performance and High-Efficiency regions are powered down. Few peripherals remain with extremely low leakage.LPTIMER, LPCMP, LPRTC, and LPGPIO
Table 3 Device Power Mode Summary

Power Management Unit

A Power Management Unit (PMU) in the processor controls the internal LDOs, DCDC, and clock generation. It is responsible for enabling or disabling the resources depending on the SoC Power Mode.

Power ModeClock AvailabilityDCDC Status
GOAll Clocks and PLL enabledDCDC: full power
STANDBYLFXO or LFRC (32,768 Hz) and
HFRC (38.4 MHz or 76.8 MHz)
DCDC: efficiency
STOPLFXO or LFRC (32,768 Hz) onlyDCDC: Off
Table 4 Power Management Unit Resources

Retention Memory – RTSS-HE TCM

For the RTSS-HE, its 512KB TCM supports an option to be retained while the device is placed in low power STOP Mode or STANDBY Mode. Normally, turning off either RTSS results in a power loss to the TCM and its contents will be erased. In the case of the RTSS-HE an option is available to keep its TCM powered on separately. Enabling this feature increases the static current of the device in the low power modes but it allows for increased flexibility in applications using the low power modes. They can benefit by keeping code or data retained in memory between sleep cycles. The tradeoff depends on multiple factors. A memory fill at every wake cycle takes time and energy if the TCM is not retained. On the other hand, increased static current should be accounted for when the TCM is retained.
Current consumption at 3.3V: 2.3 μA to retain 256KB or 4.7 μA to retain 512KB of RTSS-HE TCM

Retention Memory – Backup RAM

There is 4KB of SRAM in PD-0 which can be used for saving critical information before entering low power STANDBY or STOP Mode. This memory will stay retained and its contents will never clear if a valid voltage is maintained on the VDD_BATT supply. The SE shall firewall this memory space for security. The application CPUs may send a service request to the SE if they need to save information in the VBAT backup memory.
Current consumption at 3.3V: 65 μA

Selectable Clock Frequencies

Much of the processor runs at high frequencies centrally sourced from the device PLL. The output frequency of this PLL remains fixed and programmable clock dividers reduce the frequency for the device subsystems. Applications may choose to run at a high frequency where it is more efficient to complete a task quickly and immediately enter a sleep cycle. Other applications may not have the flexibility and must stay awake longer. These mux options let application cores run at reduced clock frequencies to minimize power consumption and extend battery life.

CPU States and Power Modes Detail


A processor subsystem is in the RUN State any time the processor core is executing instructions, including the no-operation instruction or an empty for(;;) or while(1) loop. This means power and clocks to the processor core, local memories, and local resources are all enabled. The rest of the device, shared peripherals, and shared memory are available as well. Dynamic power and static power within the processor subsystem are highest in this mode. This is shown in Figure 2 below.

Figure 2 All Processing Subsystems Are in RUN State
Figure 2 – all processing subsystems are in RUN State


Entering SLEEP State is done by a processor core executing the WFI instruction. Power is left on for the processor core and its local memories making this mode the fastest to resume operation. In the SLEEP State a processor core will clock gate, but local resources and shared resources continue operating. Local and shared resources may then generate interrupts to resume RUN State on the processor. There is no limit to interrupt availability when entering SLEEP State; all interrupt sources in PD-6 SYST will be available to wake any processing subsystem.
Dynamic power within the processor subsystem is reduced in this mode relative to the RUN State. Power consumption caused by any activity in the shared resources domain remains present.

Figure 3 All Processing Subsystems are in SLEEP state
Figure 3 all processing subsystems are in SLEEP State


If the application allows, a processor subsystem may be placed in OFF State where the processor core, local memories, and local resources are all disabled and powered off. Resuming RUN State via hardware interrupt is possible using LP- peripherals on RTSS cores as shown in Table 5 OFF State Wakeup Sources. To wakeup other subsystems a request from the RTSS can be made to the SESS. Then the SESS will be responsible for waking that core.

The example in Figure 4 below shows an example of some processing systems being OFF and some being in SLEEP, demonstrating that not all processing systems have to be in the same state.

Figure 4 all processing subsystems are in OFF state
Figure 4 all processing subsystems are in OFF State while SESS and RTSS-HP are in SLEEP State

Other subsystems may be active but dynamic power and static power within the processor subsystem are eliminated in OFF State. Optionally, at the expense of increased static power, the local memories can be retained in RTSS-HE only.

Wakeup Sources in CPU OFF State
Interrupt SourceDestination
Table 5 OFF State Wakeup Sources


When the High-Efficiency Region is on, and all higher numbered domains are off (PD-3 and up), the device is in STANDBY Mode. Using STANDBY Mode makes available some clocks and modules that are not available in STOP Mode. The cost is higher static power relative to STOP Mode. The clocks and modules enabled by this mode are in High-Efficiency Domain (PD-2) shown in the Power Domains Table. To wakeup other subsystems a request from the RTSS-HE can be made to the SESS. Then the SESS will be responsible for waking that core.
Current consumption at 3.3V: 65 μA

STANDBY Mode with all subsystems in OFF State
Figure 5 STANDBY Mode with all subsystems in OFF State

Wakeup Sources in STANDBY Mode

While in STANDBY Mode, these modules can generate interrupts to wake the RTSS-HE. Note: When the RTSS-HE wakes up during STANDBY Mode this will not automatically wake up the High-Performance Region or the Shared Resources domain (PD-6). The RTSS-HE may operate in this mode executing within its TCM alone using only the peripherals within PD-3, without using any of the device’s shared resources. It can then go back to STANDBY Mode for the next sleep cycle. If other cores or peripherals are needed, then it can make a request to the SESS to wake the system out of STANDBY Mode.

Interrupt SourceDestination
Table 6 STANDBY Mode Wakeup Sources


When the AON Region is on, and all other regions are off, the device is in STOP Mode. The infrastructure for routing interrupts to the subsystems is not available anymore. Instead, any wake requests from the AON Region will wake the SESS. From there the SESS will wake one of the processor subsystems which requested entering STOP Mode.
Current consumption at 3.0V: 1.6uA, includes 4KB Backup RAM retention

 STOP Mode with both HE and HP Regions OFF
Figure 6 STOP Mode with both HE and HP Regions OFF

Wakeup Sources in STOP Mode

A selection of LP- modules can be enabled before entering STOP Mode. These modules can then generate interrupts to wake the device. It is the application’s responsibility the configure the below modules for generating an interrupt. After which the application will request STOP Mode through a SE API call.

Interrupt SourceDestination
Table 7 STOP Mode Wakeup Sources


Maximizing the time spent in STANDBY and STOP Mode is critical and will lead to a successful low-power design. To help with this, the Ensemble processor provides timers, comparators, GPIO, and other LP-peripherals that an application can use to efficiently dictate its wake and sleep cycles. Multiple memory retention options are available so that applications requiring fast resume can save their context and then wake up as quickly as possible. Finally, when the device is awake, the high-performance features of the Ensemble family, such as the Ethos-U55 accelerator, Cortex-M55 with Helium DSP extensions, and high-frequency clock options, allow software to complete tasks quickly and return to sleep immediately.


Your request was sucessfully submitted. Our team, will get back to you as soon as possible. Thank you.