[Cryptech Tech] Alpha Avalanche Noise Block

[Павел Шатов]

On 14.04.2015 21:00, Fredrik Thulin wrote:
> On Tuesday, April 14, 2015 02:51:56 AM Павел Шатов wrote:
>> On 13.04.2015 9:55, Fredrik Thulin wrote:
>>> I looked at the SEPIC converter you suggest (LTM8045EY) and thought maybe
>>> we need to talk about that choice a bit more. It is fairly expensive
>>> (Farnell US has it at $12.27 in single quantities) and while that may be
>>> acceptable at this point (having stable HIGH_V is also of course
>>> important to us) it also seems to only be available in BGA package?
>>>
>>> I think that is worse, since it means we can't really test it before
>>> getting the Alpha boards manufactured, and also can't use it in simpler
>>> units. How did you choose that exact part? Could we reasonably find
>>> another SEPIC converter with good enough vcc-noise properties but another
>>> package?
>>
>> Fredrik, excuse me, I don't understand what simpler units you are
>> referring to.
>
> Boards such as my STM32 based ones, or stand alone noise boards for the
> Novena. It might not be that the Cryptech project is interested in building
> more of these, but others might from our designs.
>
>> Alpha board is not going to be simple, it is going to be
>> complex. Our CPU and FPGA and RAM are all BGAs. If we can't avoid BGAs,
>> let's use them to our advantage then. Alpha is not going to be cheap,
>> because Artix-7 200T costs about $300. Because of that, I didn't think
>> that $12 power regulator was like very expensive.
>
> Agree the Alpha will be neither simple nor cheap.
>
>> Two-layer boards that you can etch and solder yourself are one story,
>> Alpha board is going to be completely different story. I don't think
>> that it is appropriate to pick same components for different types of
>> boards. If you need some regulator for you experiments, we can try to
>> find something in a more DIY-friendly package.
>
> Based on transparency and audit-ability I would prefer the same components to
> be used in simple as well as advanced designs, but not if the drawbacks
> outweigh the benefits of course.
>
>>
>> Speaking of DC/DC converters, they can be divided into three general
>> types. The first type includes bare controllers. They are very small and
>> cheap, because they contain only control circuitry, they require several
>> external components (power switch, inductor, diode, capacitors). They
>> are typically used when large output current is required. Since we need
>> only about 50 mA, we don't need this type of converter.
>>
>> Converters of the second type have control circuitry and internal power
>> switches. They still require external inductor, diode and capacitors,
>> you also have to do some math to calculate required inductance value.
>> These converters require prototyping, because you must check that you've
>> selected appropriate inductor.
>>
>> The third type of converter is a regulator module. Such converters have
>> built-in inductors and diodes, so they only require input and output
>> capacitors. LTM8045 is an example. I've suggested to use it, because it
>> is straightforward to use, you don't need to prototype it. It has PCB
>> layout example in the datasheet that will 100% work from the first
>> attempt. When I recommended this module, I thought, that paying a few
>> extra bucks would be better than dealing with inductors and diodes.
>
> Thanks for the explained reasoning. I'm not used to things working 100% from
> the first attempt, but with that much confidence in the Alpha design I agree
> the objection to using another BGA is void.
>
>> Anyway, as far as I understand, Joachim is back, so you're going to have
>> a meeting with BitSim guys soon. You can ask them about this issue. Tell
>> them, that besides digital components, you have sensitive analog
>> circuit, that requires stable +12V (probably higher) and draws about 50
>> mA. Let's listen to what they say.
>
> Yes, good idea. We have a meeting scheduled before lunch on Friday 24/4.
>
> Btw, I'm running some real tests on the noise circuit in isolation to see what
> voltage is actually required, and the power requirements.
>
> Will get back with more details, but so far I've tested 12V (2.642 mA) and
> 12.5V (3.129 mA).

Speaking of voltage required for avalanche effect, you should consider 
part-to-part parameter variation. Transistors are analog components, if 
you buy ten BC818 transistors, they will all have different DC current 
gain. Because of this, if you have 10 identical boards, they may require 
slightly different voltages for avalanche effect.

--
With best regards,
Pavel Shatov