Power supplies for ADAS applications and infotainment systems
13 February 2015
The performance increase of ADAS and infotainment applications puts a growing strain on application processors, causing power consumption to rise.
Growing integration and shrinking circuit geometries lead to higher leakage currents also increasing power consumption. The efficiency of the power supply has to be top priority.
As well as the potential difficulties mentioned above, quick availability is also required. Quite often, it is necessary to power up the processor domains in a pre-determined sequence. In some cases, the processor itself and its power supply should also be scalable to suit a full product family.
Requirements and problems
Video data conditioning for infotainment systems, speech-recognition applications or the analysis of different sensors are used in ADAS applications for traffic-sign recognition, collision warning with brake assistance, collision avoidance, lane-departure warning, and parking assistance. All this puts an enormous strain on the processing performance of the application. This trend is accommodated by increasing integration, multi-core architectures for concurrent operation and dedicated video accelerators.
However, increased power consumption also results from higher transistor counts and increased leakage currents caused by the lower RDS of circuits based on smaller geometries. After all, supply voltage reduction has not kept pace with the growing degree of integration.
It is also necessary to consider the larger process variations of smaller process geometries. ‘Hot‘ silicon featuring high dynamic performance and higher leakage currents must be differentiated from ‘cold’ silicon, which is characterised by lower performance and much lower leakage currents. Without additional measures, each processor would have to be operated at a supply voltage sufficient to meet the performance requirements even with the ‘coldest’ silicon, although this entails high losses for ‘hot’ silicon. Depending on the processor load, current consumption can vary considerably between several milliamps and several amps. Power supplies therefore should provide a wide output current range and quickly respond to dynamic load steps.
Depending on the application, it may also be necessary to support a ‘sleep mode’, keeping only small sections of the processor alive while others are powered down in order to save power. On the other hand, the system must be fully operable within a very short time after a wake-up call (i.e. via the CAN bus).
The goal is to provide the various processor cores with exactly the amount of power needed to deliver the necessary processing performance. This applies to applications as well as to scalable full solutions capable of processing various amounts of sensor data and featuring different levels of graphic resolution.
Most processors also have specific requirements regarding the order and delays of their power-up and power-down sequences. Many applications must fit into very limited space (e. g. beneath a seat, in the hatch door, in the rear-view mirror or in the transmission tunnel). The challenge therefore is to provide a rapidly available power supply, which is efficient and adaptive and provides sequencing support.
The ‘Dynamic Voltage Frequency Scaling’ (DVFS) approach has been used for quite a long time to optimise the battery lifetime of mobile devices including laptop computers, tablets and mobile phones. DVFS varies the clock frequency of the processor cores depending on how much processing performance is required. Additionally, it reduces the supply voltages at lower clock frequencies, which can make a difference of more than 200mA for different operating points. Until now, this approach was rarely used in automotive applications because extensive radiation tests must be passed for each potential clock/frequency combination in order to exclude any potential interferences from other devices. The first DVFS systems now emerging in automotive applications need suitable power supplies.
Deviations resulting from the manufacturing process are another aspect. As mentioned above, ‘hot’ silicon features excellent dynamic performance at the cost of elevated leakage currents, while ‘cold’ silicon is somewhat slower with much lower leakage currents. To achieve the same processing performance, ‘cold’ silicon needs higher supply voltages. If this voltage would also be used for ‘hot’ silicon, elevated leakage currents would result in excessive losses. This is taken into account in modern power-supply solutions including Adaptive Voltage Scaling (AVS) and SmartReflex by adapting the voltage to the silicon with an accuracy of several tens of millivolts.
Although it results in additional requirements, the multitude of processor cores also provide additional potential to save energy. Cores featuring independent supplies can be powered down individually or switched into different sleep or low-power modes. However, a specific power-up and power-down sequence is required in order to avoid any cross currents. At the same time, specific start-up time requirements must also be met in automotive applications. For example, it must be ensured that the parking assistance is operable immediately after powering up the vehicle. Power supplies should also have active discharge circuits for the power-down process. Otherwise, the blocking capacitors and back-up capacitors would have to discharge very slowly across the load before the next supply voltage could be switched off.
Using high clock frequencies for the switching regulator helps to meet tight space requirements because smaller external components (especially inductors) can be chosen. As another positive side effect, additional screening and filters can be eliminated at frequencies above 1.7MHz because the spectrum is then located above the medium-wave band. On the other hand, high switching frequencies result in high switching losses. Thus, a suitable tradeoff must be found between size, emissions and power consumption. An elegant method to achieve optimised switching losses across a wide output current range, accompanied by good dynamic load response, is to use multi-phase supplies. Instead of using a single switching regulator for a specific load, several supplies are used in parallel. In a two-phase configuration, for instance, the load current is delivered by two regulators operating with a phase shift of 180°, while in a three-phase configuration three regulators operate at a phase shift of 120°. Under low-load conditions, only a single phase will be powered up, thereby eliminating any losses in the other phases.
The TPS659038/39-Q1 family from Texas Instruments is a highly flexible and highly integrated solution. The devices include up to nine switching regulators providing up to seven independent voltages and up to eleven LDO’s. Each regulator is individually configurable. The output voltage of the switching regulators can be set in 10-mV steps (in 20-mV steps above 1.6V), while the LDOs‘ output voltages can be set in 50-mV increments. With the exception of the internal supply voltage (which powers up immediately after applying the external supply), each supply voltage can be arbitrarily placed within the sequence. Therefore, it is possible to use the devices for many different processor families of the OMAP5, TDA2x or DRA7xx series and for common application processors including their peripherals. Depending on individual requirements, each regulator can be assigned to a specific processor domain. Furthermore, the supply voltage can be set according to the operating point, and sleep modes can be activated if necessary. In most cases, the devices will provide additional resources for supplying external memory (DDR2, DDR3, DDR3LV, NOR/NAND Flash, etc.).
If the on-chip regulators should not be sufficient, dedicated outputs or GPIO lines can be used to power additional external sources which can be integrated into the sequence and the sleep modes if necessary.
In order to achieve high efficiency across a wide load range and a good dynamic load response even without large output capacitors, the device supports multi-phase operation. While it operates in the single-phase mode under low-load conditions, one or even two phases can be added powering the same output with a phase shift of 180 or 120°. Thus, an efficiency of approximately 90% can be achieved between 100mA and a maximum of 9A with a switching frequency of 2.2MHz (depending on the input and output voltages). Due to the high switching frequency, very small 1µH inductors are sufficient. The multi-phase approach also yields a typical dynamic load-step response of 3% of the output voltage, even with output capacitors with a capacity of only 47µF. Taking into account the integrated power transistors, this solution is a compact, highly effective package.
To meet the sequencing and power-up time requirements, the desired sequence is pre-programmed into a non-volatile memory so it can be retrieved for the power-up and power-down processes. Depending on the resources and delay times, all outputs can be powered up within 10 to 20msec. The outputs of all switching regulators and linear regulators feature active discharge circuits enabling quick transitions into the sleep modes and rapid power-down processes.
A variety of internal settings can be read or manipulated using the I2C or SPI interface, which is particularly useful for output voltage adjustments in AVS (adapting to the silicon) or DVFS applications (to match the operating point of the specific processor domain). Thus, scalability can be ensured even without modifying the boot voltages. The power supply can be reused whenever a solution must be downscaled or when changing to a solution featuring higher performance.
With its internal, selectable interrupt handling and its integrated ADC whose results can be used to trigger an interrupt (if the temperature is rising, for example), the device can often be used for all the housekeeping functions as a replacement for a separate microcontroller. Functionality is completed by additional features like the integration of GPIOs into the sequence as inputs or outputs, a watchdog and different synchronisation options.
Combining efficiency, sequencing support, EMI limitation, flexibility, and minimum space requirements, the TPS659038/39-Q1 PMIC family represents an ideal power supply for processors from the DRA7xx, TDA2x and OMAP5 families and similar processors from other manufacturers. Thanks to its scalability and compatibility, various configurations can be implemented using this PMIC family.
Contact Details and Archive...