Investor Engineering Discussions

[mongo]

Yes, the amount of Silicon in the anode is not 100% clear. but battery day flags a clear increase.

When we saw the anode foil at Berlin, that looked dark, indicating at least some Graphite...

When in doubt I always take battery day literally, often I find people have a bias towards a more 'normal' rather than 'literal' interpretation of battery day.

Doh, yeah I think my brain added an "all" in there.

 

[Buckminster]

 

[mongo]

Um. That's a little odd.
ADC's come in various "bit" sizes: 8 bit, 12 bit, 16 bit, etc. They'll typically have a precision reference voltage (band-gap diode reference or something) hooked up to the top of a resistor chain and ground or something at the bottom of said chain. An 8-bit ADC can measure 256 possible values, plus or minus the usual inaccuracy. But 8-bit ADCs are relatively uncommon; the cost difference between an 8-bit, 12-bit, or 16-bit ADC is minimal, so people usually go for the 12-bit (4096 values) or 16-bit (65536 values). The bigger bits are slower, typically, than the smaller bit types, but we're talking typical sample rates of 1 MSa/s and up for the cheapies, and I'm pretty sure that a BMS doesn't need to know state of charge on a microsecond-by-microsecond basis. (Now, if you want to sample along at 10 GSa/s, then you might be restricted to an 8-bit ADC, but that's a different story.)
Further, I suspect you're not going to find very many people willing to sell you an ADC that takes in 800V directly; that's a good way to end up with silicon vapor. Typically, one uses a resistor divider so the maximum input voltage to the ADC is 80% of the maximum for the ADC. (And the accuracy of the measurement is directly related to the accuracy of the resistors; in an application like this, one would need to use matched resistors, built on the same substrate, around .1% accuracy, typically, without spending too much money.)
So, to measure 800 V, we'd divide this down to 80% of the maximum of the ADC, then measure. So, using a 12-bit ADC, we'd have 800/(4096*0.8) = 244 mV/step; with a 16-bit ADC, that'd be 800/(65536*0.8) = 15 mV/step.
First off: in my extremely uneducated BMS opinion, 244 mV or 122 mV (for a 400V system) seems too large to keep track of the voltage on a battery chain; for sure, if one was monitoring the overall battery voltage, 16-bit and 15mV (or 7.5 mV for a 400V system) seems to be the way to go. Now, if I can figure this out without an envelope, so can anybody else.
Second: I dunno if Porsche (or Tesla, or whomever) is running every single battery cell in series (unlikely in the extreme) or all in parallel (impossible if one wants high voltage), or some combination of series and parallel (a lot more likely), the individual voltages that the BMS would be monitoring would be individual cell voltages or smaller groups of cells, then do some math to add it up to the overall battery pack voltage (if desired). Remember: They're trying for the amount of charge on cells, I would think. But that means that the voltages they're measuring would be, for the Porsche, 800/N, where N is the number of series groups.
Which means that my step sizes up above would also be divided by N. So, for example, there's 10 bunches of cells in series, with 80V across each bunch, then a 12-bit ADC on that bunch would have 244/10 = 24.4 mV. Which is probably getting down to the point where a 12-bit ADC would be doable.
Conclusion: This ain't rocket science, at least not for a practicing EE. So, while it sounds like a nifty idea (ADC mistake!), it's probably not that. Sorry.

Yeah, BMS doesn't measure the whole pack. Either individual group independently (diff amp to ADC) or referenced to module ground based and math to isolate each group. LTC6802-1 for example with 12 bit ADC.
Taycan modules are 12 cells in a 2p6s setup, so 25ish volts max * 33 modules = 825V max.

 

[Tronguy]

Yeah, BMS doesn't measure the whole pack. Either individual group independently (diff amp to ADC) or referenced to module ground based and math to isolate each group. LTC6802-1 for example with 12 bit ADC.
Taycan modules are 12 cells in a 2p6s setup, so 25ish volts max * 33 modules = 825V max.

Yep, sounds very right. Which means with a 12-bit ADC we got (244 mv/step)/33 = 7.3 mV/step or so.
As a wild guess, I'd think that they'd run the ADC right on top of the module using the local ground, then communicate with the ADC via optos and I2C/SPI; they make chips for that very purpose.

 

[mongo]

Yep, sounds very right. Which means with a 12-bit ADC we got (244 mv/step)/33 = 7.3 mV/step or so.
As a wild guess, I'd think that they'd run the ADC right on top of the module using the local ground, then communicate with the ADC via optos and I2C/SPI; they make chips for that very purpose.

Powering the BMS depends on how they deal with low voltage. Externally power won't drain cells forever. Internally makes the module more stand alone. Chipsets do run isolated interconnects commonly in a daisy chain setup. More SPI than I2C to simply things since there are only two nodes and single direction data flow.

 

[mongo]

Um. I agree with you about cutting blocks of cells in/out of a group in parallel. But I have a quibble about voltages and such and where they come from.

I've formed an opinion (probably wrong, but who knows?) that the actual battery charger in all these BEVs is some form of DC-DC converter, running at $DIETY's own switching frequency. Input side would be DC from the car's on-board AC->DC converter or direct DC from a charger. Output side would be the DC voltage and current used to charge the battery. These things are highly efficient in their power conversion, especially with relatively high voltages on input and output.
I mess with these things from time to time at work on a regular basis, although admittedly the outputs are used to power up silicon, not batteries. Having said that, though: I haven't run into one of these in the past decade or two that doesn't have the capability of measuring both its output voltage and current; and, even if one had some kind of home-brew circuit that didn't have the output V and I monitoring built in, it's truly trivial to add that after the fact; Hall-effect current sensors are out there and are cheap, not to mention just measuring the voltage across a low-value resistor.

In principle, then, an electron shouldn't be able to fall in the woods without somebody/something noticing.

I suspect that the problem with Porsche's BMS, assuming the whistleblower is onto something, has nothing to do with basic monitoring. Since I got the Tesla, I've been reading the odd papers here and there posted online about different battery chemistries, longevities, anode and cathode structures, and all that. I'm not an expert, by any means, but the algorithms used to charge/discharge the batteries in a BEV look to be very, very challenging, and, if one gets those wrong, then, yeah, the whistleblower may be onto something.

Porsche has a separate DC-DC to support <800V charge stations. 50kW default, 150kW optional upgrade.
Tesla and everyone else takes DC charging straight to the pack. Since these charge events usually cut off at 90% or so, the high SOC low current performance is less of an issue.
Household AC charging is single unit AC to pack DC on most vehicles. Given the issue is reportedly occuring at lower charge rates, I'm inclined to think there is an issue with the charger not properly limiting the output current, possibly with a BMS interaction (since the system should terminate charging on an overvoltage event). It could even be the charger ripple is too high so peak voltage is excessive , but doesn't show up in the filtered measurements.

 

[Tronguy]

Porsche has a separate DC-DC to support <800V charge stations. 50kW default, 150kW optional upgrade.
Tesla and everyone else takes DC charging straight to the pack. Since these charge events usually cut off at 90% or so, the high SOC low current performance is less of an issue.
Household AC charging is single unit AC to pack DC on most vehicles. Given the issue is reportedly occuring at lower charge rates, I'm inclined to think there is an issue with the charger not properly limiting the output current, possibly with a BMS interaction (since the system should terminate charging on an overvoltage event). It could even be the charger ripple is too high so peak voltage is excessive , but doesn't show up in the filtered measurements.

Far be it from me to argue. In my experience, getting things 100% right on the first, second, or even the third try can be a challenge. Which makes Porsche's, "We don't need your help" response one of particular idiocy. I've done serious technology transfer; "Lessons Learned and What Went Wrong" always wakes up the masses in the class and leads to a lot fewer problems down the road. For some reason, the receiving participants in these kinds of classes always start off with, "Yeah, yeah, simple, simple, I'm bored." and ends with people sitting on the edges of their seats, wide eyes,, with, "And THEN what happened!!!???" questions. Especially if one attaches dollar figures to the issues. And if the product being transferred is making serious bucks. And the participants are the ones who're going to have to carry the ball, later.
Fun.

 

[MC3OZ]

That is not required. A structural LFP battery for the 3/Y can be made with prismatic cells.

Do you mean the BYD Blade battery?

A agree that the pack is structural but I'm not sure how much the prismatic cell contribute to the structure.

I'm not doubting this statement, just puzzled, and requesting additional information.

I agree that a prismatic LFP battery can fit into the Model 3 made at Fremont and probably the Model Y.

I do doubt that a prismatic LFP battery is a god "drop in" replacement for the structural battery pack in the Model Y.

I see Tesla making LFP 4680 cells as being depended on one than and one thing only, whether Tesla has sufficient Lithium.
It would need to be after around April 2022) form memory after patents and license fees expire.

I see Tesla 4680 LFP production as partially related to whether the Lithium clay extraction process works and the timeline for that process to achieve volume production. Piedmont Lithium is a different matter entirely, Tesla may have replace it with supplies form Australia for now.

I still see Tesla 4680 production at Austin starting with Nickel -Manganese as previously advised.

I agree the Cybertruck production process is gaining momentum, but I don't think we have seen a 8000 series casting machine installed at Austin. Without that machine, I doubt they can make Cybertruck.

More generally it seems that the buildings for Cybertruck construction are generally complete and equipment is being installed. it will take time for all of that equipment to be installed and calibrated for parts supply chians to be organised.

My conclusion, if Cybertruck production does start early it will be very low volume, it will not impact on cell volumes for Model Y production in any meaningful way.

 

[MC3OZ]

4680 capacity will be needed for more volatile energy dense chemistry for long range and performance vehicles, LFP doesn't need the same protection and can be made structural in prismatic formats.

To make LFP 4680 cells Tesla needs Lithium, equipment and staff, the other raw materials are relatively easy to get.

Also it is possible to ask CATL to make LFP in 4680 format.

Prismatic LFP is also good fit for energy storage.

A agree nickel-manganese and nickel, for Model Y, Cybertruck and Semi are a higher priority.

 

[MC3OZ]

My additional thoughts are.

1. Tesla plans to make 3 TWh of cells by 2030, there is a good chance at least 1.5 TWh of these cells are LFP. Some modest "proof of concept" production of 4680 LFP in 202/2023 seems desirable.
2. Importing LFP form China cancels out some of the cost advantages of LFP, There is little doubt if LFP cells could be made in the US / EU from locally sourced raw materials, those cells would be cheaper than Chinese imports.
3. CATL can make 4680 LFP for use in Chinese made cars.
4. Standardising all vehicles on the 4680 cell format with a mix of chemistries is the most efficient model.

 

[petit_bateau]

A question for those more in the loop than I am. In the review of the Tesla 12V management boards the current measurement was done using nifty mini shunts. In the BMS discussions some of you are noting that Hall effects are being used. What is driving the different selection decisions ? Is it the need for isolation ? or cost ? or min power loss ? or availability ? or what ? Are there any usage trends ?

 

[mongo]

A question for those more in the loop than I am. In the review of the Tesla 12V management boards the current measurement was done using nifty mini shunts. In the BMS discussions some of you are noting that Hall effects are being used. What is driving the different selection decisions ? Is it the need for isolation ? or cost ? or min power loss ? or availability ? or what ? Are there any usage trends ?

Mini shunts can be very accurate but there is a trade off of signal level and power loss. Hall effect are also fairly accurate but can be influenced by external magnetic fields. The handy thing with Hall current sensors is that they provide electrical isolation to the circuit being measured. So a normal 5V micro can read the current on a 1kV DC line with no additional circuitry needed. Wheras, with a shunt, you are ground referenced to keep the sensed voltage low (which does put a voltage offset on the ground), or high side and your circuitry is dealing with the high voltage and needs to translate it down to microcontroller levels.

12V sensing is less onerous and a lot of the high side smart FET parts I've used have a built in current sense output that interfaces easily with micros.

 

[petit_bateau]

Mini shunts can be very accurate but there is a trade off of signal level and power loss. Hall effect are also fairly accurate but can be influenced by external magnetic fields. The handy thing with Hall current sensors is that they provide electrical isolation to the circuit being measured. So a normal 5V micro can read the current on a 1kV DC line with no additional circuitry needed. Wheras, with a shunt, you are ground referenced to keep the sensed voltage low (which does put a voltage offset on the ground), or high side and your circuitry is dealing with the high voltage and needs to translate it down to microcontroller levels.

12V sensing is less onerous and a lot of the high side smart FET parts I've used have a built in current sense output that interfaces easily with micros.

Thanks. Brings me more up to date.

(Hall effects were sufficiently expensive when I was last involved in this stuff, not so many years ago, that we worked quite hard to avoid them.)

 

[mongo]

Thanks. Brings me more up to date.

(Hall effects were sufficiently expensive when I was last involved in this stuff, not so many years ago, that we worked quite hard to avoid them.)

Yeah, Halls do cost more, but super handy when dealing with 50+ amps with externally facing rails. Also helps with low power modes vs analog circuitry that may have leakage paths.

Watched the Ingineerix video. Tesla is using precision high side shunts with standard FETs to allow exact control of the overcurrent thresholds along with system monitoring with bidirectional control.

Drop on busbars are a neat idea. Less overhead than one piece custom parts.

 

[Tronguy]

Yeah, Halls do cost more, but super handy when dealing with 50+ amps with externally facing rails. Also helps with low power modes vs analog circuitry that may have leakage paths.

Watched the Ingineerix video. Tesla is using precision high side shunts with standard FETs to allow exact control of the overcurrent thresholds along with system monitoring with bidirectional control.

Drop on busbars are a neat idea. Less overhead than one piece custom parts.

I mainly mess with Hall effect transducers on 48V isolated battery voltages. It's a rather cool idea: the magnetic field created by the current being measured plays across a fixed-current square area, making the electrons curve around, and changing the voltage across the square area.
Trick is: The whole business is temperature sensitive, reference sensitive, and has some 1st and 2nd order variations on the zero points and all. The new devices I've seen are a buck or so or less and have all the compensation built right in, at least in the 0-10A range that I play in.
Shunts tend to be more accurate, since what one is measuring is the voltage across a resistor, and a low-value resistor can be ridiculously accurate. There are chips that are rail-to-rail, or at least rail on the high side, with fixed gain, that provide easy access to a reasonable voltage level for an ADC somewhere. But then one has to worry about Vos and such, since said devices are basically op amps. All good, clean fun.

 

[mongo]

Predicting the direction of Cyber Truck drive and suspension can help anticipate stock price changes. Here are some observations that may help.

Tesla has invested a lot in motors, but to obtains power per lb the RPM has to be high. (Look at the routers at the local hardware store). If the motor moves to the the wheel, the gear reduction system would go with it. Pot holes would not be a friend.

That leaves half shafts and a central motor (and gear reduction) to accommodate pot holes. When I shopped for half shafts, there was a limit on drive angle at each end. In other words, significant suspension travel required significantly long half shafts.

I think so far these are just facts. Maybe someone else can predict. Or maybe there are other facts. Hoping someone can build off this.

With the air suspension most of the time the half shaft will be fairly aligned, but we've alreafy seen the issues on original models with Ludicrous launch induced CV failure. If needed, they could connect the half shafts to the far side of the drive unit with one shaft fore and one aft. Easily done with dual motor drive units, but creates an additional constant angle offset.

 

[mongo]

While all that is true if you're looking for max towing capability you'll want the largest pack and I doubt that will be offered in the lowest priced RWD version if such a thing is offered. Even when towing having AWD offers advantages over RWD in many situations. Towing boats and utility trailers are probably the most common uses and both will often be in low traction situations such as boat ramps and dirt roads/construction sites. Percieved value of AWD/4WD over RWD is high and reflected in the price of products. All things considered I see the likelihood of a RWD dual motor truck from Tesla to be low.

Tis an interesting question on product mix.
Agree is it all moot if Tesla doesn't make a RWD version. The reservation mix highly favored AWD trims, but those were also tied to range numbers, so a longer range RWD might be popular. Especially for people not in snowy regions.

If RWD did exist:
Would Tesla, having already developed a dual motor rear drive unit, then create a single motor variant to pair with a single motor front drive unit?
Even ignoring R&D cost, AWD with two motors would be more expensive than RWD with two motors.
RWD with one rotor would be cheaper than dual, but likely harder to package due to increased power. It would probably more more efficient though (mass and drag reduction) which helps with a lower cost pack.

With 3 pack sizes and 2 drive unit part numbers, Tesla can cover a lot of use cases. MR may be too much granularity, depends on step size from SR to LR, whether SR could possibly be LFP.
SR RWD: lowest cost
SR AWD: low cost and all weather (Winter commuter)
MR RWD: Longer trips with towing (weekend boat hauler)
MR AWD: Same, but winterized
LR RWD: Long range towing version (range may be better than AWD, but trading weight for regen)
LR AWD: Highest power, winter resistant

 

[JRP3]

Would Tesla, having already developed a dual motor rear drive unit, then create a single motor variant to pair with a single motor front drive unit?

Even ignoring R&D cost, AWD with two motors would be more expensive than RWD with two motors.

Tesla already has heavy dual motor vehicles with AWD, the S and X. They have plenty of power and they aren't load bearing so should be strong enough. Cost and development would seem to be almost nil. Take a bit off the top end, gear them a bit lower, maybe tweak the software, and bolt them right in.

 

[mongo]

Tesla already has heavy dual motor vehicles with AWD, the S and X. They have plenty of power and they aren't load bearing so should be strong enough. Cost and development would seem to be almost nil. Take a bit off the top end, gear them a bit lower, maybe tweak the software, and bolt them right in.

Yeah, making single or dual motor drive units is not a huge endeavor. I'm may be over commonizing to reduce unique part #s and supporting engineering needed (torque vector vs open differential)

 

[JRP3]

Somewhat related: Tesla officially brings closure to the Model Y's forgotten variant