This may be the most rigorous article you can find discussing SM8550.
Please indicate the author when reposting.
Background#
The author obtained a SM8550 (Honor Magic5Pro) motherboard.
So they opened it.
The product is a pop package (Package on Package).
TOP Package#
The top package mark is:
2WB77 D8CCP
According to the search:
The manufacturer's code is MT62F1536M64D8CL-023 WT,
indicating that the chip is LPDDR5 96G 1.5GX64 FBGA 8DP.
So they decapped it.
It is known that the D8CCP chip is actually a product of Samsung SEC.
SEC stands for Samsung Electronics Company.
It's strange.
Bottom Package#
After opening the top package, they moved on to the bottom package.
This time, they obtained the SM8550, also known as 8Gen2.
They didn't provide detailed performance analysis, as it can be easily found.
This time, they obtained the mass production stage SM8550,
specifically the SM8550 002-AB.
The complete code is Qualcomm-SM8550-002AB-FX215S2Y.
Top Mark Analysis#
SM stands for Snapdragon Mobile.
8550 is the product model, which belongs to the SM 8 series, 5th generation, or 8gen2.
002-AB is a variant.
FX215S2Y is the top mark code, in the format of xx/x xx/xxx.
The first xx represents the manufacturer and packaging factory.
The second x xx represents the Year and Week.
The third xxx represents the traceability code.
F stands for TSMC production, X stands for Amkor, Japan packaging.
2 represents the year 2022, 15 represents the 15th week.
S2Y is the traceability code, but its specific meaning is unknown.
These are the data that can be obtained from the top mark.
Bottom Package Decap#
Next, they decapped the bottom package.
Unfortunately, it was damaged.
In order to save some data, they tried to increase the magnification.
However, the magnification was not enough, so they increased it further.
To facilitate viewing, they switched to a better die.
The die mark is located at the center of the image.
Qualcomm HG11-31145-2
Its specific meaning is unclear.
Marked area.
The topmost layer is the fanout metal layer.
Good Die#
The reason why deeper areas cannot be seen is because of the metal layer blocking, making it impossible to see the poly layer.
Measurement results:
Die size is 11.20mm x 10.57mm, which is 118.384mm2.
It is known to have 17 billion transistors, resulting in an average density of 143.6004MTr, which matches the TSMC N4 density.
The actual number of transistors is calculated to be 170.117808 billion.
Process Analysis (SEM)#
Although it is known that this is TSMC N4, it needs to be verified using SEM (scanning electron microscope).
Process Analysis#
The analysis here is simple.
This is TSMC's technology roadmap.
It can be seen that N4 is a branch of N5, inheriting N5's 6TUHD/7.5THD libraries.
Specific data:
Fin Pitch:
Yield Analysis#
To calculate the yield, the dpw formula is used.
Based on N5 D0 calculation:
When D0=0.75, the yield is 43.99%, or 238 out of 540 dies.
There is also thermal design power simulation, but it is a rough calculation and does not require the same analysis as Hi36a0 for capacity/yield analysis.
(From ISSCC 2023 / SESSION 2 / DIGITAL PROCESSORS / 2.2)
Dieshot Layout#
Next is the acid etching and layout drawing.
Because the dieshots they obtained were low-resolution and compressed, they decided to create their own high-resolution dieshot.
They completed the layout drawing.
Then they extracted individual images.
This is the size of the CPU Cluster+L3+SLC.
They extracted the CPU image.
This is a comparison of the core size of X3 A710 A715 A510 A510Cluster. It can be clearly seen that Sdm is actually a 5-cluster design, rather than 143 or 1223. For more details, refer to an 8gen2 smartphone, as the author does not have one.
GPU#
This is the GPU of the 8G1+ series. It can be clearly seen that it is a 2-cluster structure, with each cluster having 4 cores (2CU).
The characteristic of Adreno is that its Tmus are not located within the Alus Core, so the labeling is approximate.
Compared to Adreno730, Adreno740 has the advantage of an additional cluster, resulting in 12 cores (6CU 3 clusters).
The number of Alus reaches 1536, and the frequency is labeled incorrectly, but the actual frequency is 680Mhz. The theoretical performance is 1536x0.68Ghz x2 = 2088.9Gflops.
This is a comparison of the GPU size of this generation and the previous generation. The smallest one is actually the GPU of SM8475. Please measure the specific sizes yourself, as the author hasn't learned how to do it yet.
Modem#
The modem part of X70 is not significantly different from the previous generation X65.
There may be some minor changes, but the author hasn't studied basebands.
Size comparison:
The baseband of D9200 may be M80, but the author is not sure.
ISP#
Qualcomm always has a 3-core ISP every year, but they couldn't find any information about the ISP, so they made a rough estimation.
Strange Unit#
This is a strange unit. It is known that the top part is the internal memory, but the function of the bottom part is unclear. In this generation, the area has increased by 1.5-1.7x, which is strange.
Theoretical Performance#
Detailed performance tests can be found in many media sources, so the author didn't include them in the article.
Disclaimer#
Please indicate the author as @Kurnal when reposting.
Feel free to repost the images, indicating the author as @Kurnal.
The dieshot was created because the existing images were low-resolution.
The original article is available at https://kurnal.xlog.app/SM8550, while the others are reposts or reposted by the author.
Thank you, Alipay v50.