Thread: Gamma processor type

Thanks for that datasheet fasulye, I didn't have that.

From my brief look at Gamma code it clearly utilises both 3DES and 512 bit RSA, calling des_ecb_crypt() and AEmodN_for_RSA() from hsoc01.c. The RSA certainly isn't a software emulation, indeed we see in hsoc01.h that the RSA unit is addressed at $d20000-d20270. As expected those addresses are shown as "unimplemented" in the MM2107 datasheet. That doesn't mean Gamma isn't using crypto, rather it means Gamma isn't using MM2107.

Your photo is interesting, well done for producing it, it's much clearer than I could have obtained!

However I maintain that Gamma isn't using MMC2107 but an enhanced version with crypto accelerator. Plenty of chips nowadays have RSA, the silicon requirement is small and cheap ( in 44-pin package gives you 512 bit RSA via SPI interface in 63mS for under $4 in 1-off quantity!). There's no question that the whole reason Irdeto switched to new platform for their V4.1 cards was to get fast RSA, and Gamma required the same. You sure as hell can't do it in real time without hardware acceleration.

Looking at the MMC2107 datasheet, I think you may be right that a bootloader was programmed at the factory before the dies were mounted in their cards. I was surprised about that but it now seems probable. I doubt the factory was programming the actual Irdeto emulation code, for fairly obvious reasons, but rather a generic loader that Gamma team used to inject their application later.

One thing that puzzles me is how they managed the supply decoupling. I assume they wired all the Vss pads together when mounting on the die, and all the Vdd together. But parts like this with onchip PLL are usually quite fussy about decoupling and prefer individual caps close to each Vss and Vdd. Obviously that's not an option for smartcard packaging unless integrated decoupling capacitors are used, which I've not seen. I was half expecting the die power routing to have been reworked for smartcard packaging. What power connections did you observe on the die?

The other interesting thing about the power supply is the MMC2107 datasheet says it's designed for 2.7-3.6V. Yet I know plenty of people that run Gammas quite happily at 5V 6MHz without overheating, for certain. That further supports my view that the Gammas aren't actual MMC2107s but a derivative.

And although this hasn't been discussed much, I suspect that there may be both 3V and 5V versions of Gamma card in existence. That sounds a bit hard to believe, I know, but some people found their Gammas definitely overheated unless they reduced supply to 3V, while others run fine at 5V. I heard rumours a long time ago of "later batches" of cards being produced to "solve the overheating problem". Those rumours might have been rubbish, but it's possible that early version of Gamma card was strictly 3V only but demand for cards that operate in 5V equipment (as most CAMs are) may have driven development of a 5V tolerant version.

Who knows. Perhaps one day the story will be told, with enough detail to lay all the speculation to rest. In the meantime it's fun to guess what was done back then in those secret factories

You are correct with your theory about the gamma being a descendant of the MMC2107. Freescale never produced a m-core based platform with an embedded hardware Cryptography. The die picture is complete and there are no other interconnected dies or passive components.

It is strongly believed that the gamma team commissioned the development of a custom solution.

Originally Posted by fasulye

It is strongly believed that the gamma team commissioned the development of a custom solution.

That seems probable. But the question that raises is, what on earth possessed them to choose MCore rather than, say, ARM (as )?

If I were approaching a VLSI design house to produce a custom crypto smartcard, knowing that powerful pay tv companies would soon be trying to identify me, my philosophy would be:

1. Choose an independent manufacturer who isn't contracted to major pay tv companies and who is beyond the reach of their subpoenas.
2. Give them the feature/performance requirement (eg 8K SRAM, 100mS 512bit RSA, 3DES, RNG, 32-64K code flash, 2K NV data eeprom/flash, ISO7816 voltage & current constraints etc) but don't over-specify the implementation.
3. Let them leverage the silicon libraries and expertise they already have, to minimise cost, development time and risk of teething problems.

So the question is, who would choose MCore? Why would a VLSI house end up using MCore in the Gamma design when others like ARM are more widely used? Especially since there are several ARM impementations around with integrated crypto? The only possible reasons I can think of:

• The house had successfully produced custom smartcards in the past, and of those designs the MCore one had the highest performance. Changing architecture without good reason would have been a needless risk.
• The house couldn't or didn't want to license ARM for some reason, they'd identified crypto integration as the biggest risk, and they'd had more experience extending MCore than any other processor library, so using MCore was the safest way to mitigate the risk.
• The Gamma had come across an MCore eval kit in his past for some reason, it was an embedded system he felt comfortable with, and so he (inappropriately) mentioned it in his functional specification document. The VLSI house took it as a design requirement without question.
• The VLSI design house pays royalties on its libraries and MCore worked out cheaper than alternatives.
• Maybe MCore offers some technical advantage over alternatives that I've missed?

Only the first of those seems likely to me. What do others think?

Brilliant post guys.
Now I know why I fix TVs lol

Anonymity and secrecy would have been the driving constraint used in the selection of the VLSI house. Most like an existing design was reused or modified as the basis for the gamma. Hence why the die has the approximately 100 outer pads.

Originally Posted by gw1

From my brief look at Gamma code it clearly utilises both 3DES and 512 bit RSA, calling des_ecb_crypt() and AEmodN_for_RSA() from hsoc01.c. The RSA certainly isn't a software emulation, indeed we see in hsoc01.h that the RSA unit is addressed at -d20270. As expected those addresses are shown as "unimplemented" in the MM2107 datasheet. That doesn't mean Gamma isn't using crypto, rather it means Gamma isn't using MM2107.

The DES, on the other hand, appears to probably be a software implementation. Does ? I wonder if it's vulnerable to power analysis...

Originally Posted by gw1

So the question is, who would choose MCore? Why would a VLSI house end up using MCore in the Gamma design when others like ARM are more widely used?

Because other cores are more widely used?