agx_sram_mW_mV.txt

100 414.0
200 414.0
300 415.0
310 415.0
320 415.0
330 416.0
340 416.0
350 417.0
360 417.0
370 419.0
380 420.0
390 420.0
400 422.0
410 422.0
420 422.0
430 422.0
440 422.0
450 422.0
460 422.0
470 423.0
480 425.0
490 425.0
500 426.0
510 429.0
520 429.0
530 430.0
540 430.0
550 432.0
560 434.0
570 434.0
580 437.0
590 437.0
600 439.0
610 441.0
620 442.0
630 443.0
640 446.0
650 449.0
660 449.0
670 451.0
680 456.0
690 457.0
700 460.0
710 464.0
720 466.0
730 469.0
740 472.0
750 476.0
760 480.0
770 484.0
780 488.0
790 492.0
800 499.0
810 500.0
820 508.0
830 512.0
840 517.0
850 524.0
860 529.0
870 536.0
880 544.0
890 552.0
900 560.0
910 567.0
920 576.0
930 586.0
940 594.0
950 606.0
960 616.0
970 629.0
980 638.0
990 651.0
1000 666.0
1010 679.0
1020 693.0
1030 709.0
1040 724.0
1050 742.0
1060 760.0
1070 778.0
1080 797.0
1090 820.0
1100 841.0
1110 864.0
1120 890.0
1130 914.0
1140 942.0
1150 969.0
1160 1000.0
1170 1031.0
1180 1064.0
1190 1099.0
1200 1136.0
1210 1174.0
1220 1214.0
1230 1258.0
1240 1302.0
1250 1350.0
1260 1399.0
1270 1452.0
1280 1504.0
1290 1564.0
1300 1625.0
1310 1686.0
1320 1755.0
1330 1824.0
1340 1899.0
1350 1975.0
1360 2059.0
1370 2145.0
1380 2234.0
1390 2328.0
1400 2428.0
1410 2534.0
1420 2644.0
1430 2760.0
1440 2882.0
1450 3007.0
1460 3144.0
1470 3282.0
1480 3432.0
1490 3587.0
1500 3750.0
1510 3924.0
1520 4104.0
1530 4294.0
1540 4494.0
1550 4703.0
1560 4920.0
1570 5152.0
1580 5396.0
1590 5649.0
1600 5917.0
1610 6198.0
1620 6494.0
1630 6804.0
1640 7130.0
1650 7469.0
1660 7830.0
1670 8206.0
1680 8603.0
1690 9017.0
1700 9453.0
1710 9912.0
1720 10392.0
1730 10898.0
1740 11428.0
1750 11984.0
1760 12568.0
1770 13180.0
1780 13824.0
1790 14498.0
1800 15207.0
1810 15953.0
1820 16734.0

notes.md

The raw data is in the format millivolts milliwatts. Milliwatts is rounded to integer (probably floor). It is the sum of 8 independent calculations with different scaling constants that add up to 130, which are rounded independently, but the inner function is known to be independent of that constant. So you can treat it as a single calculation, with an 8 mW error range for the values (since it is rounded 8 times before adding). There is a constant offset that adds up to around 414 mW (plus error) due to core power which I cannot eliminate from the testing matrix.

$V_{core} = 100\ \mathrm{mV}$

$N_{clusters} = 8$

$E_{quant} = \displaystyle \sum_{k=1}^{N_{clusters}}U_{k} \cdot 1\ \mathrm{mW}$ (Irwin–Hall distribution, $n=N_{clusters}$)

$P_{core} \approx 414 \ \mathrm{mW} + E_{quant}$

$k_{leak} = 130$ (this is a known factor of the output, but there may be others)

Current is probably calculated separately from voltage, so the expected form is:

$P(v) = P_{core} + V \cdot k_{leak} \cdot f(V)$

Equivalently,

milliwatts = 414 + millivolts * 130 * f(millivolts / 1000)

Your job is to find $f(V)$. If you can get all the output errors to within a +8mW range (or less), you win! ✨