Part-to-Part Variability of Total Power

Variability in the manufacturing process of the silicon die leads to variability in most electrical parameters of the processor. The total power consumed on the die follows this rule and some variations can be observed from part to part when they are placed in identical software and hardware conditions. This variation is both software and hardware dependent and it should therefore be characterized in real case conditions to estimate the expected maximum when in production. To do so, it is recommended to build at least 20x PCBs and measure their individual power and junction temperature during the application-specific use case. From this measurement set, classical statistical tools (average and standard deviation) can be used to assess the production maximum.

The following table provides the results of measurements carried out on 20x SAMA7G54-EK evaluation kits when running an iPerf(1) test at fCPU = 1 GHz, with TA = 70°C.

  1. 1.iPerf is a Linux network benchmark tool. See Use Case 2: iPerf Test.
Table 1. Measurement Results on 20x SAMA7G54-EK Boards
  Total Power Junction Temperature
Average (μ) 738 mW 90°C
Standard deviation (σ) 34 mW 0.92°C
Estimated maximum (μ + 3 x σ) 840 mW 92.8°C