Apple looking to ditch Broadcom in addition to Qualcomm

[Cmaier]

Apple developing custom WiFi, Bluetooth, and 5G chips for iPhone

Apple has been working on developing its own modem for several years, looking to reduce its reliance on Qualcomm. A...

9to5mac.com

Apple to Start Transitioning Away From Qualcomm Modem Chips in Late 2024 or 2025

Apple wants to start replacing the Qualcomm modem chips that it uses in iPhones as soon as 2024, according to a new report from Bloomberg. Apple has...

www.macrumors.com

Apple developing custom WiFi, Bluetooth, and 5G chips for iPhone

Apple has been working on developing its own modem for several years, looking to reduce its reliance on Qualcomm. A...

9to5mac.com

Apple looking to drop Broadcom as wi-fi, bluetooth, cellular supplier | AppleInsider

Apple has been planning for some time to ease its dependence on Broadcom, and it appears that it may happen by late 2024.

appleinsider.com

The complexity of wifi/cellular/bluetooth is a lot lower than the complexity of the CPUs that Apple is already successfully designing, at least for the digital portion of the chip. The analog portions are beyond my ability to judge difficulty. I thought this would be the year Apple would try it, but it looks like maybe next year is the year.

I do hope they have gotten good at this sort of design - having owned different phones with either qualcomm or Intel-based radios, the intel-based ones are always a little worse.

One note: when they are talking about combining wifi/blutooth/5g on one component, that raises interesting possibilities for Macs.

 

[mr_roboto]

The complexity of wifi/cellular/bluetooth is a lot lower than the complexity of the CPUs that Apple is already successfully designing, at least for the digital portion of the chip. The analog portions are beyond my ability to judge difficulty. I thought this would be the year Apple would try it, but it looks like maybe next year is the year.

I do hope they have gotten good at this sort of design - having owned different phones with either qualcomm or Intel-based radios, the intel-based ones are always a little worse.

One note: when they are talking about combining wifi/blutooth/5g on one component, that raises interesting possibilities for Macs.

Are you counting the DSP as part of the analog design? My impression is that this is where most of the magic is. But like you I don't have direct experience.

I share your trepidations, Intel's modem group never was able to match Qualcomm's. Hopefully it was mostly a resource issue (far more extensive testing and tuning, etc) and now that they're with Apple they're being given everything they need to compete head to head.

And yeah, would be neat to get a cell modem in every Mac!

 

[Citysnaps]

I wouldn’t go so far to say I’m skeptical of success. Maybe a little bit. Hopefully Apple has a handful of communications and signal processing systems engineers with a lot of direct experience in that area. If that’s covered, implementation should be easy.

I have a ton of respect for Qualcomm and its deep well regarding modern digital communications systems and theory. And of the founders of the company who propelled the company forward long ago; particularly Jacobs and Viterbi.

Some of the challenge is interpreting the different wireless interface standards requirements created by standards bodies. Sometimes that involves cleverness in interpretation, resulting in a favorable (in viability and cost) outcome that might not have crossed the minds of competing customers.

Years ago my company marketed our digital down-converters (digital radio), digital up-converters (digital transmitter), digital filters, QAM modulators/demodulators, etc to most of world’s cellular telecom companies. This was mostly for the infrastructure side; ie basestations or BTSs. A European customer of ours was looking at one our chips with respect to spurious free dynamic range and out-of-band filter rejection requirements. They had a novel interpretation of the standards spec and came up with filter tap weights to create the filter to meet the spec. We programmed the filter in our device, took some data, and sent the results back. The customer was pleased. Not knowing or being advised that was proprietary information, we published the tap weights and filter response in an updated data sheet and app note. The customer found out after getting a new data sheet and was pretty upset.

A long story short…there’s a lot of systems engineering secret sauce from engineers with loads of experience in that field sweating fractions of a dB in performance that drives success.

 

[Cmaier]

Are you counting the DSP as part of the analog design? My impression is that this is where most of the magic is. But like you I don't have direct experience.

I share your trepidations, Intel's modem group never was able to match Qualcomm's. Hopefully it was mostly a resource issue (far more extensive testing and tuning, etc) and now that they're with Apple they're being given everything they need to compete head to head.

And yeah, would be neat to get a cell modem in every Mac!

DSP is digital, and is not all that complicated to design, once you have the equations right. Getting the equations right is what Qualcomm is great at.

 

[dada_dave]

I know very little about this field but what I’ve heard previously is that in addition to whatever intrinsic technical challenges there are to designing such chips, it’s the additional layer of designing a good chip that doesn’t step on someone’s (usually Qualcomm or Broadcom) substantial body of patents. Basically little point in designing your own chips if you’re simply going to up paying lots in royalties anyway and/or end up with something not as good. I can’t judge for myself the veracity of that, but this seems to get written about every time it’s brought up.

 

[Cmaier]

I know very little about this field but what I’ve heard previously is that in addition to whatever intrinsic technical challenges there are to designing such chips, it’s the additional layer of designing a good chip that doesn’t step on someone’s (usually Qualcomm or Broadcom) substantial body of patents. Basically little point in designing your own chips if you’re simply going to up paying lots in royalties anyway and/or end up with something not as good. I can’t judge for myself the veracity of that, but this seems to get written about every time it’s brought up.

The benefit to apple is differentiation. By combining wifi, bluetooth, satellite, and 5g in a single chip, for example, they can reduce power tremendously. By combining all that with their apple silicon (GPU/CPU) they can get even more potential benefit. They can also do things like embed H2-style communication, ultra wideband, etc. They can fab on better processes than Qualcomm if they want. There are just a lot of potential technical advantages. A lot of risk, sure, but also advantages. And the more they do this and build up the team, the more input they will have into future standards, the more standards essential patents that they will own, and the better deal they will be able to get from qualcomm, Broadcom, etc.

 

[mr_roboto]

DSP is digital, and is not all that complicated to design, once you have the equations right. Getting the equations right is what Qualcomm is great at.

Yeah I really meant the equations when I typed "DSP". None of the digital circuit design should be even slightly hard, other than maybe going the extra mile on power optimization (as Apple is wont to do).

I know very little about this field but what I’ve heard previously is that in addition to whatever intrinsic technical challenges there are to designing such chips, it’s the additional layer of designing a good chip that doesn’t step on someone’s (usually Qualcomm or Broadcom) substantial body of patents. Basically little point in designing your own chips if you’re simply going to up paying lots in royalties anyway and/or end up with something not as good. I can’t judge for myself the veracity of that, but this seems to get written about every time it’s brought up.

Because of this thread I googled up Apple's press release announcing their agreement to acquire Intel's mobile RF modem group, and it talked quite a bit about patents. Enough that you can tell the personnel weren't the only reason for Apple to acquire that business unit.

But it's always what comes after the acquisition that's important. How well did they integrate the team, how much brain drain happened, and how well did Intel's product plans adapt to Apple's internal needs. It has been taking them a fair bit of time to ship something.

 

[Cmaier]

Yeah I really meant the equations when I typed "DSP". None of the digital circuit design should be even slightly hard, other than maybe going the extra mile on power optimization (as Apple is wont to do).

I‘ve only been involved in the technology as part of my current career (where I’ve learned a lot about the standards for bluetooth, LTE, 5G, and wifi), and while the terminology is crazy (a lot of 3-letter acronyms, a lot of calling things by different words than computer engineering even though it’s talking about the same concepts, etc.), the basic elements are not too complicated (viterbi, puncturing, trellis coding, QAM, PN offsets, etc. Etc.). But doing it all in a way to be as efficient as possible and as resistant to noise and errors as possible seems like it’s quite an art.

 

[dada_dave]

The benefit to apple is differentiation. By combining wifi, bluetooth, satellite, and 5g in a single chip, for example, they can reduce power tremendously. By combining all that with their apple silicon (GPU/CPU) they can get even more potential benefit. They can also do things like embed H2-style communication, ultra wideband, etc. They can fab on better processes than Qualcomm if they want. There are just a lot of potential technical advantages. A lot of risk, sure, but also advantages. And the more they do this and build up the team, the more input they will have into future standards, the more standards essential patents that they will own, and the better deal they will be able to get from qualcomm, Broadcom, etc.

Aye I understand the end goal, it’s just that people keep bringing this up every time Apple doesn’t ship something in this space or why the Intel modem was never quite as good. I know you’ll have an opinion on why Intel never got it right .

Also, I can’t fully remember as this was awhile ago, but someone mentioned once there might be downsides to integrating everything together, I think it was heat? At least with respect to smartphones, spreading the modem out from the CPU/GPU keeps each cooler as in a smartphone it’s simply harder to passively cool both when they’re both working hard. Basically it was a caveat to the idea that full integration wouldn’t have trade offs of its own. I think it was an Anandtech comments section - it would be hard to find and I can’t remember if people agreed with that idea or there was significant pushback. Obviously Apple thinks it’s worth it and on this part no doubt they’ve done the work beyond napkin math that such an approach would yield the desired benefits.

 

[Cmaier]

Aye I understand the end goal, it’s just that people keep bringing this up every time Apple doesn’t ship something in this space or why the Intel modem was never quite as good. I know you’ll have an opinion on why Intel never got it right .

Also, I can’t fully remember as this was awhile ago, but someone mentioned once there might be downsides to integrating everything together, I think it was heat? At least with respect to smartphones, spreading the modem out from the CPU/GPU keeps each cooler as in a smartphone it’s simply harder to passively cool both when they’re both working hard. Basically it was a caveat to the idea that full integration wouldn’t have trade offs of its own. I think it was an Anandtech comments section - it would be hard to find and I can’t remember if people agreed with that idea or there was significant pushback. Obviously Apple thinks it’s worth it and on this part no doubt they’ve done the work beyond napkin math that such an approach would yield the desired benefits.

You want the analog stuff separated from the digital stuff, both for heat and electrical noise. But separate chiplets in a single SoC package is good enough.

 

[mr_roboto]

You want the analog stuff separated from the digital stuff, both for heat and electrical noise. But separate chiplets in a single SoC package is good enough.

Furthermore, the analog designers I've talked to have told me there's no real benefit to process nodes newer than... I think 45nm? It's been a while. Apparently, analog-optimized transistors haven't scaled down to any significant degree since then.

So, that just makes it more attractive to push analog mixed signal blocks into their own special chiplet. As long as there's not too much digital in them, you can get away with building the thing in a much older and cheaper node.

 

[leman]

I remember reading an opinion by someone who was allegedly familiar with modem design, a they said that this is much more complex than one thinks. Too many little details and unwritten rules. I am not surprised that Apple keeps delaying their modem. Most likely they had a working chip for a while but would experience weird malfunctions when operating in the field.

It’s a bit similar to the problem of supporting external displays (it’s not enough to implement the spec correctly - you have to know which popular displays do not implement the spec correctly and adapt to that!). But with modems it’s probably even more complicated.

 

[Cmaier]

I remember reading an opinion by someone who was allegedly familiar with modem design, a they said that this is much more complex than one thinks. Too many little details and unwritten rules. I am not surprised that Apple keeps delaying their modem. Most likely they had a working chip for a while but would experience weird malfunctions when operating in the field.

It’s a bit similar to the problem of supporting external displays (it’s not enough to implement the spec correctly - you have to know which popular displays do not implement the spec correctly and adapt to that!). But with modems it’s probably even more complicated.

I’ve seen the schematics for some of these things, and they are far simpler than CPU stuff. That said, the verification problem is big, and it takes years of knowledge to build up test suites. And there’s a lot of analog stuff that I imagine is very tricky to get right, even if all the logic is perfect.

 

[Citysnaps]

I remember reading an opinion by someone who was allegedly familiar with modem design, a they said that this is much more complex than one thinks. Too many little details and unwritten rules.

That's my belief as well. And that's on the communication/signal-processing systems engineering side - not on hardware/chip implementation.

Also...The small company I worked for at one time discussed developing similar communications system modem chips (not for cellular handsets). We had the communications systems engineering expertise and chip design experience to do that. In the end it was Qualcomm's many thousands of patents and patent applications that deterred us. Even if we didn't step on some, the threat and cost of defending a potential lawsuit made that a no-go.