电池供电的边缘 AI 传感：案例研究实现低功耗、始终开启的功能.pdf

1、Battery-Powered Edge AI Sensing:A Case Study Implementing Low-Power,Always-On CapabilityPeter FennDirector,Advanced Applications GroupAvnet Designing for low power consumption Edge AI design goals,constraints and applications “Always on”smart sensors and AI/ML processing Wireless connectivity low-po

2、wer techniques Device security in low-power use-cases Case study:Sensors,NPU,Secure-MCU,Wi-Fi/BT+more Revisit of design trade-offs and lessons learned References and resources for more infoAgenda2Designing for low power consumptionBattery-powered IoT/Edge devices need to be energy-efficient!Average

3、Current Consumption is a key specification(i.e.the“area under the energy curve”)Power =V x IEnergy=V x I x Time=V x mAhWhat needs to be powered-on and when.?Always-onInterruptActivatedPeriodicReportingSmart SensorsYAI ProcessingYHost MCUYYWirelessYBattery energy needed to achieve the applications fu

4、nctionality should be analyzed over time!Answer design questions like these:What sensor sample-rate is needed?How much processing latency is acceptable?Can components sleep between events?3Designing for low power consumptionLow Power design techniques include the use of High-value pull-up/pulldown r

5、esistor values High-efficiency PMIC/voltage-regulator devices DVFS(Dynamic Voltage&Frequency Scaling)(dynamic power is proportional to the voltage squared!)Minimizing the ON time“duty-cycle”of devices FET load-switches to turn-off non-active circuits High-side FET“Load Switch”4Low Power design techn

6、iques(continued)Reduced rail voltage for power-hungry peripherals Clock gating(and/or reducing clock frequencies)Strapping unused digital inputs(dont let them float)Taking advantage of MCU&wireless low-power modes Prevention*of phantom power and latch-up(*use a buffer with power-down isolation for s