Google Tensor dieshot
Image from Techinsights , marked by me.
Total density:
| size(mm^2) | MB | MB/mm^2 |
| 5.196721532 | 16.5 | 3.17507873 |
| MB | size(mm^2) | density | |
| X1 L2 2MB | 2 | 0.637221296 | 3.138627055 |
| L3 | 2 | 0.632872744 | 3.160192976 |
| LLC | 8 | 2.509719416 | 3.187607328 |
| A76 L2 | 0.5 | 0.161472401 | 3.096504407 |
| GPU L2 | 4 | 1.245759426 | 3.210892822 |
Compare to other:
| A14(TSMC N5) | M1(TSMC N5) | Google Tensor(SS N5) |
| 3.527017114 | 3.338617945 | 3.17507873 |
| Kirin 990 5g(TSMC 7nm EUV) | Kirin980(TSMC 7nm) | Lakefield(Intel 10nmSF) |
| 2.11467507 | 2.287187327 | 1.675987565 |
Detail:
| size(mm^2) | |
| X1 core | 2.51 |
| A76 core | 0.93 |
| GPU total | 18.1 |
| GPU core | 0.773 |
| Google TPU total | 8.51 |
| Google TPU Systolic array | 0.4044 |
Its looks like Samsung N5 are very close to TSMC N5 in real density.SRAM Block density are close,and A76 core are way smaller than TSMC N7s.
Google tensor have 108mm^2 of size.
