iPhone 5 - Virtual Teardown Guessing
Don Scansen posted on September 20, 2012 |
For anyone hoping that the Apple iPhone 5 media launch would provide clarity, it seems it opened the door to more questions. These unknowns also opened the door for many pundits to evermore unresearched and unsupported speculation - to the point that we are getting component cost estimates before analysts have looked inside the phone. To would-be customers and the authors of these reports, I pose several questions. First, if you estimate $167.50 before looking at the device and its internals, and iFixit's inevitable teardowns reveal errors in your assumptions about component manufacturer, type or configuration, will you be tempted to fudge the numbers to match what you had before? To avoid any appearance of impropriety in this regard, I call on the cost estimators to provide at least a general understanding of the major assumptions used to arrive at their calculated manufacturing cost for the Apple iPhone 5 generated prior to getting access to the hardware.

If this information is not forthcoming, pre-teardown estimators run the risk of the bloggers and mainstream media who have reported these numbers deciding such calculations were invalid. Were they based on anything tangible? Is it possible the dollar values given to some items were complete guesses? Numbers that just "sounded right?" So what is the value of buying an expensive teardown-BOM report? Can't someone with a casual interest in technology spend a couple hours on the Internet and arrive at the same point?

With respect to the cost numbers assigned to key components, I would be hard pressed to argue against guesses. There is so little known about the system design and major new components like the A6 processor chip which is the first or second most expensive item regardless of the costing methodology. The UBM TechInsights estimate actually pegged the Apple-designed processor as the most costly item, and it might cost more than the display. But this isn't The Price is Right. The suggestion is that they have an idea just how much more expensive the A6 is.

Furthermore, costing for Apple is a different animal than the conventional approach used for products that use standard off-the-shelf components. It is all well and good to pull apart a Google or Kinkdle tablet, look up the part numbers, guess the volume, and look up a price at Avnet or DigiKey. I once argued that guessing a system cost before looking inside was reasonable for more standard platforms where the general design was known and high-level details like the processor make and model, amount of flash and screen specifications were known. But Apple is not repackaging a reference design in its own skin.

And there is a second level of detail where this "sight unseen" approach to costing fails. Apple in select territory as a consumer electronics company using its own processor design and alone in creating designs captive to its own products. How do you go about determining the cost for a device that is not on the open market? In fact few of its specs will be known even after an earnest reverse engineering effort.

The first step is to determine what few thing we know about the A6 processor. Apple announced that the new die would be 22% smaller than the A5 found in the iPhone 4S. Apple bills the new processor as twice as fast while maintaining comparable power consumption to the older device. But what can this tell you about the cost? A very simplistic approach that might be favoured would be to look only at the wafer manufacturing cost since the A5 die size is known, and therefore, so is the A6.

Wafer foundries charge by the wafer to produce integrated circuits. For a given process complexity - number of interconnect levels, minimum feature size etc. - one need only know how many die (again with certain assumptions as to yield for the manufacturer) each wafer can accommodate. Again, there needs to be an assumption here. Is the A6 manufactured in the same foundry at the same process node as A5? If you have analzyed the chip, there are excellent tools to provide the cost, especially the IC Knowledge Cost Model.

Interestingly, the TechInsights A6 cost estimate pegged the upcoming processor at $28 versus only $21 for the A5. Clearly, their cost analysis is more complex. But how? There is more to the cost of this chip, but again, so much more that we do not know.

We do not know the package construction and if and how the DRAM is packaged with the device. Based on the approach used by Apple in iPhones compared to iPads, we can assume that the DRAM will be found within the processor package rather than in discrete packaged DRAM chips as used in conjunction with the A5X on the iPad 3. But will the A6 use conventional package-on-package (P-o-P) that has been in use since the A4 launched? There was some speculation that A6 would be moving to TSMC primarily to take advantage of the leading IC foundries production-ready through-silicon via (TSV) process. The TechInsights blog about the BOM results even acknowledges this.

In a TSV design, the chip becomes a true 3D circuit with interconnection between stacked chips in vertical through holes through the silicon die. The interconnections are shorter than conventional approaches that interconnect die by external bond wires or conductors within the package materials. In the existing P-o-P approach, the processor has landing pads on its lower surface to connect to the motherboard as well as on its upper surface to accept a standard DRAM package configuration. Putting the two together has most in common with population of a circuit board. But heat build-up can be an issue.

I have no reason to confirm new processor packaging rumors except to say that iPhone 5 designers would probably welcome the additional cooling since the TSV conductors are well known to provide a good thermally conductive path out of the chip. The TSV design facilitates adding a more effective metal heat sink on top of the system package. For instance, the A5X includes a metal heat sink since there is no P-o-P configuration for the DRAM. Apple did say the A6 runs twice as fast. Is part of the reason improved cooling? (Incidentally, new benchmark results possibly from review units suggest the overall performance is better than 2X. The iPhone 5 numbers are also a little embarrassing for some of the Android competitors.)

To understand where the A6 might fall, one should look at the evolution of the Apple A(x) processor line. If you were trying to understand the difference in prices, you should certainly make such an effort. Paul Boldt has provided as good a perspective on this as is currently available. Of course, this may change once teardowns begin and the die photos are published by Chipworks and TechInsights that reveal major architectural features of the chip. But we will cross that bridge on the weekend.

If you made it this far, you are doubtless happy the following is my final comment on the A6 cost analysis. How much did the chip design cost? What type of chip IP was purchased (ARM etc.)? What type of licensing deal did Apple negotiate? How much did the internal design team cost (salaries, workstations, software licenses)? There are good tools for assessing the manufacturing side like IC Knowledge, but again, we need more of that information than we have. Of course, Nvidia, Qualcomm, Samsung and TI are in the same boat, but their product marketing factor these costs against the volumes they expect to ship when setting the selling price. Apple has no selling price for the A6.

There are other system-level assumptions that would impact the accuracy of a non-teardwon cost estimate. Flash memory is a big one. Apple is the largest consumer of the non-volatile memory chips and very much sets the spot price. Guessing how much Apple may have paid for this commodity is an art in itself. Apple procurement people are serious about this. It is tough to know when the prices were negotiated - possibly well in advance of taking the first shipments.

But there is yet another assumption built into this. Will Apple continue to use standard NAND flash disks with built in controller (like just having your complete USB memory stick on the board)? Or will they push controller logic functions onto the A6 design? Remember anobit? The timing of the sale could not affect the A6 design, but Apple was doubtless buying anobit flash controller IP well before the acquisition - just as they did with the Intrinsity Hummingbird speed enhanced custom ARM processor design on the A4. By the way, NAND flash memories require a lot of smarts to take care of where, when and how to write data, but this is a detailed technical discussion better left to another day.

What do we make of system cost estimators and the buzz they generate? The secret sauce to these estimates and their believability is PR and media connections. Place the quotes in the right space, and most people will believe them. It doesn`t hurt to have the media and analysis arms under one corporate umbrella. Would the WSJ really use less than credible sources? Of course not. Just believe the number. But don`t believe me. I'm just jealous of the attention.

Recommended For You