Post a Comment
Inside Apple's iPhone: More than just a dial tone
Prying Eyes examines the engineering inside history's most hyped cell phone. See which processors, memory devices, and other ICs Apple's engineers selected, and ponder the architecture decisions that hint at future iPhone generations.
By Brian Dipert, Senior Technical Editor -- EDN, 7/27/2007
I admit it: I'm a bit of a wimp when it comes to Prying Eyes teardowns. Get me a piece of gear with an easily (or at least somewhat easily) removable top, side (or top, depending on how you orient it) or back panel, and I'll gleefully grab a screwdriver regardless of the widget's cost. Well…within reason, that is, and especially if it's no longer working, anyway. If it's free or free-after-rebate, I'm not worried about breaking it during dissection. An already partially disassembled patient is a bonus, as is cereal on the side. Even a double-digit pricetag isn't too much of a barrier, either to my nerves or to the accountants who approve my expense reports.
| Image gallery
Click any picture to enlarge. You'll also find links to the full-size images embedded within the text of the article. All images courtesy of Portelligent. |
|
|
|
 Digital PCB (side 1) |
 Digital PCB (side 2) |
 RF PCB (side 1) |
 RF PCB (side 2) |
 Touchscreen harness |
 Touchscreen LCD |
 Docking harness (side 1) |
 Docking harness (side 2) |
 Earpiece harness (side 1) |
 Earpiece harness (side 2) |
 User-interface harness (side 1) |
 User-interface harness (side 2) |
|
But when a Prying Eyes candidate's cost crosses the triple-digit threshold, I take serious pause before pursuing the project. And when the cash register rings up a McKinley (or more), I know where to go—in pursuit of a partnership with someone else who's already tackled the teardown!
That's why I locked arms with Llamma.com for my recent analysis of Microsoft's Xbox 360 Elite. And that's why I joined up with Portelligent for this Prying Eyes project. Portelligent published a bill-of-materials inventory and a cost-estimate report soon after Apple and AT&T began selling the iPhone on the evening of June 29. I'm profoundly grateful to Portelligent for agreeing to join with me on this Prying Eyes adventure. By leveraging their existing work, I'll be able to preserve my iPhone's functionality for future hands-on reports, as Apple and AT&T evolve the device's feature set.
The information that follows comes predominantly from my discussions with Portelligent, along with published reports of their findings. I've supplemented it with my analysis of similar (and oft-contradictory) reports from iSuppli and Semiconductor Insights, coupled with additional close-up internal images published by AnandTech, Ars Technica, the FCC, iFixit, PDAParts, and Think Secret. And finally, I've added to the blend some information that I obtained from anonymous in-the-know confidants. Apple's an extremely paranoid company, both with respect to its employees and its suppliers, so I'm not able to reveal my sources' identities or even their employers.
If you peruse AnandTech, iFixit, or Think Secret's pictorial step-by-step disassemblies, or PDAParts' or Semiconductor Insights' abbreviated YouTube clips, you'll hopefully get a sense of just how precisely designed the iPhone is. For that reason, I doubt there's a large degree of automated assembly involved; the iPhone is likely constructed largely by hand in China (where iPods are built today) or some other location with low-cost, highly skilled labor. Compare the iPhone imagery to that of the Compaq iPaq 3835, which was the subject of my Nov. 23, 2006 Prying Eyes piece. Now consider that the iPhone integrates Wi-Fi, Bluetooth, and GSM cellular data and voice, along with a position sensor, none of which the iPaq provides, along with a much more powerful CPU and substantially more nonvolatile and volatile memory. Finally, realize that the iPhone's form factor is almost 3× smaller in total volume than that of the iPaq 3835 (in fairness, I should point out that the iPaq 3835's screen is 0.25-in. wider than that of the iPhone, although the two LCD's lengths are identical). After digesting this data, you'll likely walk away quite impressed, as I was, with the accomplishments of Apple's electrical and mechanical engineering teams, as well as with the integration accomplishments of Apple's various semiconductor suppliers.
At the bottom left corner of the overview photo, you'll see a 2-megapixel Micron CMOS sensor, which combines with a three-element, fixed-focus, fixed-focal-length lens to create the iPhone's flash-deficient and still-image capture-only camera.
The 3.7V lithium-ion battery in the bottom right corner of the photo consumes a dominant percentage of the iPhone's total volume, and is a key determinant of the device's substantial specified battery life. Apple estimates that after 400 recharge cycles, the battery will still retain up to 80% of its original capacity; replacing it will cost you $79 plus $6.95 shipping, with an estimated three-business-day turnaround. For an additional fee, the company will even loan you a spare iPhone while yours is in the shop.
Between the iPhone's camera and battery, in the photo, you'll find a single flex circuit that encompasses all of the iPhone's antennas; a combined 2.4-GHz Bluetooth/Wi-Fi element and a separate (and much larger) multiband (850-, 900-, 1800- and 1900-MHz) cellular element. The planar antennas attach to the phone speaker's acoustical chamber.
The 3.5-in. diagonal (2×3-in.) LTPS TFT LCD (Picture), which includes National Semiconductor's Mobile Pixel Link serial interface receiver in a chip-on-glass implementation, delivers a 480×320-pixel (half-VGA) resolution at 160 dpi, but Apple doesn't specify the display's color depth (either native or dithered). The touchscreen elements arrange in a coordinate system capable of supporting both self- and mutual-capacitance modes, thereby enabling simultaneous multitouch support managed by an Apple-labeled, Broadcom-supplied touchscreen controller and an NXP Semiconductor ARM7TDMI processor. The "How Stuff Works" site has a good summary of how the iPhone's touchscreen works its magic.
The main board assembly, swathed in EMI-shielding material, is a two-PCB (digital and RF) sandwich, which you can see between the camera module and antenna in the overview photo. Note that in the images that follow, the background grid dimension is 1 cm. One side of the digital board is dominated by a 4- or 8-Gbyte NAND flash chip (depending on the iPhone model), on the PCB's left side. If you visit Samsung's website, the largest device you'll find listed there is a 32-Gbit (4-Gbyte) NAND flash memory, and only single bit-per-cell devices are represented. While Samsung could get to 8 Gbytes in a single package by implementing a two-die arrangement, I've long suspected (and other teardowns have confirmed) that Apple is an early customer for Samsung's nascent MLC (multilevel-cell, aka two-bit-per-cell) NAND-flash-memory program. This is the same NAND flash memory used in the iPod nano.
The other dominant IC on that same side of the digital PCB is an Apple-labeled applications processor on the right, with a volatile memory underneath it in a two-package stack. Die-level analysis by Portelligent and Semiconductor Insights has confirmed that the applications processor is a Samsung design, thereby following in the footsteps of Samsung's first CPU design win with Apple in the second-generation iPod nano. A specific part number is unknown at this time; the IC may not (yet?) be available in the open market. However, hackers' analysis of the iPhone's recovery firmware has uncovered some interesting CPU specifications:
- ARM1176JZF with TrustZone, believed to be running at upwards of 600 MHz
- ARM Intelligent Energy Manager
- 16-kbyte/16-kbyte code/data cache
- Vector floating point coprocessor
- ARM Jazelle-enabled for embedded-Java execution
- SIMD high performance integer CPU with an eight-stage pipeline, capable of 675 Dhrystones/sec and 2.1 MIPS
- 0.45 mW/MHz power draw (with cache).
Firmware decoding also points to the Samsung CPU's inclusion of an Imagination Technologies-licensed graphics core, which if true would be quite ironic given my conversation with a company representative during January's Consumer Electronics Show.
What about that "volatile memory" I mentioned? It's a source of disagreement between Semiconductor Insights and Portelligent. Semiconductor Insight's video claims that the chip is an SRAM, but Portelligent is adamant that its die inspection identifies the device as a mobile (that is, low-power) DDR SDRAM.
Asked about this memory device and another I'll discuss a bit later, Don Stroud, Portelligent's director of business development, said, "We have reviewed both of the memory chips in question and still believe our assessments are correct….We did a detailed review of the package marking and then extracted the die from their respective packages to get the die markings. So, package and die markings, coupled with data sheets, make us very confident of our assessments. It would be interesting to pose the same questions to others who have analyzed the iPhone, and to know what manner of verification they employed to derive their conclusions." Translation: Touché, Semiconductor Insights!
I haven't seen the die plot myself, so I can't say for sure what memory technology's inside, but the cost-per-bit advantage of DRAM over SRAM is indisputable. And given the greater-than-50% gross profit margin that Apple's reportedly accruing, controlling bill-of-materials costs is obviously critical to the company. With that said, you can't overlook DRAM's higher average power consumption than SRAM, due to the need for periodic refresh, although low power-tuned SRAMs tend not to have significantly faster access times (random or sequential) than their DRAM peers. A feature article published nearly four years ago goes into DRAM-versus-SRAM tradeoffs in great detail, and I commend it to your inspection if you want to learn about this topic in greater depth.
In the upper right corner of the NAND flash memory is STMicroelectronics' MEMS-based accelerometer IC, used to detect the iPhone's orientation of the moment (portrait versus landscape) in order to adjust the display. Right now the function is only enabled for Web browsing, photo viewing, and iPod audio-and-video playback. As I wrote in a recent hands-on analysis, the accelerometer only reliably detects portrait-versus-landscape orientation when a user holds the iPhone at least 45° relative to the ground (and better yet nearly vertical, at ~90° to the horizontal). At the time I did my initial report, I wasn't sure if this quirk represented a fundamental sensor hardware limitation or only a constraint of the currently shipping software. However, subsequent feedback from one of my anonymous contacts suggests that the former is the case:
When the phone is held in front of you (vertically), gravitational forces on the accelerometer's "fingers" allow the phone to understand whether you are holding it in portrait or landscape mode. When you lay the phone on a table or hold it horizontally, gravitational forces operate uniformly around the accelerometer so it has no way of knowing where you are relative to it and can't determine whether you are trying to look at a portrait- or landscape-mode orientation.
Mind you, it's not a big deal; I naturally tend to cradle the iPhone cupped in my hand at waist-level, parallel to the ground, among other reasons because it requires less effort than holding the unit vertical in front of my face. In my preferred viewing position, the sensor is ineffective, but I can always temporarily raise the phone to a vertical position before rotating it, and then lower the unit again to its usual position. In fact, one could even paint this attribute as a "feature"—a sort of display-lock capability. Regardless, it's something for Apple and its accelerometer partners(s) to look at for the second-generation iPhone design.
Other primary ICs that also inhabit that same side of the digital PCB include a Wolfson Microelectronics audio codec (the same one as in the iPod nano, according to Semiconductor Insights) above the ARM CPU, power-management ICs from Linear Technology (below the ARM CPU) and NXP (the Apple-labeled IC in the bottom left corner of the PCB, according to Portelligent), a National Semiconductor Mobile Pixel Link transmitter between the CPU and NAND-flash chip, and a Texas Instruments LCD boost converter between the accelerometer and MPL transmitter. The digital board's other side is dominated by the SIM card connector, but it also houses a diminutive 1-Mbyte parallel-interface SST flash memory, which probably acts as the ARM CPU's boot device.
The iPhone's RF board is similarly jam-packed with ICs, this time concentrated on one side of the PCB (albeit with discrete components scattered across both sides). Chips you'll find on it include an Infineon GSM/EDGE RF transceiver on the right side of the PCB, a companion Skyworks power amplifier (lower-right corner), an Epcos transmit/receive duplexer (lower left corner), a Marvell 802.11b/g Wi-Fi transceiver on the left side of the PCB, a Bluetooth transceiver from Cambridge Silicon Radio just above the Skyworks IC, an Infineon cellular-baseband processor in the PCB's upper left corner, and an Intel-labeled single-package, two-die memory stack in the upper right corner of the PCB. This final device contains volatile memory and 32 Mbits of NOR flash memory. Ah yes, there's that "volatile memory" term again. As before, Semiconductor Insights says that it's SRAM, whereas Portelligent insists that it's DRAM, specifically PSRAM (self-refresh, low-power DRAM with an SRAM-like system interface).
For completeness, I'll wrap up by showing you some of the other images Portelligent provided (they don't have much semiconductor content, but they'll likely still be of interest to at least some of you):
- The front and back sides of the docking connector assembly wiring harness
- The front and back sides of the wiring harness containing the iPhone's ambient light and proximity sensors; the former controls the LCD backlight and the latter disables the display and touchscreen when you place the iPhone close to your ear.
- The front and back sides of the wiring harness containing controls for power on/off, volume up/down, and ringer mute, along with a headphone/microphone jack and the vibration unit.
Three aspects of the iPhone's design are particularly intriguing to me, aside from my general admiration of its elegance and high degree of integration:
- The NAND-versus-NOR memory-architecture argument has been a long one, waged in public forums and private customer meeting rooms alike. Apple's stance on the debate, at least as exemplified on this device, is "both." As with a PC's HDD, NAND-flash memory stores the majority of the Mac OS X operating system code (which Apple has recompiled and otherwise adapted to run on ARM; if there's any doubt in your mind, this link will likely dash your disbelief), along with music tracks, video clips, and other nonvolatile stored data. System-boot code, as with a PC's BIOS chip, resides on NOR-flash memory. In fact, the iPhone has two distinct NOR-flash devices, one each on the digital and RF boards, respectively feeding the ARM applications processor and the Infineon baseband processor.
- Speaking of ARM, its presence in the iPhone is quite impressive. Count the number of ARM processors in the device, both standalone (two) and integrated as cores within other chips; the 802.11 transceiver, perhaps, or the Infineon baseband processor. And what of the Bluetooth transceiver? See here for more ARM-everywhere system architecture thoughts from EDN's Ron Wilson.
- And speaking of multiple processors per system, here's an area where the iPhone is very unlike a PC (at least until recently, with the emergence of the GPU as a credible general-purpose coprocessor). Multiple CPU cores in the iPhone subdivide the total task and tackle different roles. The system can selectively slow-clock them or even power them off when not in use, to save power. Contrast this system architecture approach with Intel's past ARM-based XScale efforts, wherein the company (unsuccessfully, by and large) advocated a single fast CPU core for all system functions. More generally, note that whereas the ARM cores do contain hardware-acceleration logic blocks (for Java processing, for example, which the iPhone doesn't currently support…hmmm) the iPhone is fundamentally a software-fueled machine. Apple's engineers seem to have made the conscious choice not to generally rely on hardwired circuits, which arguably deliver higher performance at lower power consumption than their software-based counterparts, but at the critical tradeoff of greatly diminished current (and future) function flexibility.
