After nearly two years -- and much anticipation -- Texas Instruments has introduced the OMAP3530, the next evolution of its widely successful OMAP family of mobile application processors, which includes the OMAP2420, used in a number of mobile consumer applications including the Nokia N93 and N95. Semiconductor Insights (a division of TechInsights, which publishes EE Times) has performed a preliminary analysis to compare the new chip to its predecessor.
The OMAP platform is designed for mobile applications, including mobile phones, GPS systems and laptop computers. The chips offer a variety of features integrated into a single device. There are four separate blocks within the die that can be controlled independently, allowing parallel processing for applications to run simultaneously. This also enables blocks to be powered down when not in use to conserve battery power. The four main blocks are an ARM core, a graphics accelerator, an imaging and video accelerator (IVA) and a digital signal processor (DSP).
The OMAP2420 was designed using a 90-nm copper CMOS process. This resulted in a die size of 74mm2. In contrast, the OMAP3530 used a more advanced 65nm copper CMOS process. This allows for a reduction of almost 20 percent in total space with a final die size of only 60mm2, while offering an improvement in the functionality of the device.
The package had to change to account for the improved functionality of the OMAP3530. While the OMAP2420 was found in a 12mm x 12mm 325-pin ball grid array (BGA), the OMAP3530 was housed in a 12mm x 12mm 515-pin BGA. An interesting feature of the packaging is the design of the flip chip bonding. Most manufacturers use solder bumps distributed over the surface of the die to make electrical connections to the interposer of the BGA. In typical flip chip mounting, the die is placed with the solder bumps in contact with the lands of the interposer and then subjected to a thermal cycle to allow the solder to flow. In contrast, the OMAP3530 uses gold balls that appear similar to a wire bonding technique without wires.
The functionality of the two devices is similar, with the OMAP3530 providing an evolution to the solution rather than the revolution that the OMAP2420 provided when it was introduced. However, this does not imply that the OMAP3530 is in any way inferior to other devices on the market. The OMAP2420 just offered so much when it entered the market that it will take time to determine what else is needed.
One of the biggest changes was the ARM core. In the OMAP2420, Texas Instruments integrated an ARM1136 core running at 330MHz. While this was one of the more advanced cores available at the time, it was also being used in other baseband and application processors from other vendors. The OMAP3530 takes a new spin on the core and includes an ARM Cortex-A8 core operating at 600MHz. The ARM core is a relatively new processor IP block that has access to a 256KB. The calculated area exceeds that claimed by ARM (<4 mm2 in a 65nm process) and, therefore, it is likely that the block also includes an embedded trace macrocell (ETM) and the NEON coprocessor. The previous ARM1136 core was 5mm2, while the new ARM Cortex-A8 core is 8.9mm2.
This is the first time Semiconductor Insights has encountered the Cortex-A8 core, which is ARM's superscalar processor. The core uses NEON technology for multimedia and signal processing. It has a Jazelle real time compilation target (RCT) for execution of code languages like Java. It is capable of enhanced code density and performance with the Thumb-2 technology. Finally, it supports TrustZone for advanced security.
Both devices can support 2D and 3D graphics acceleration. In the OMAP2420, the device was able to handle up to two million polygons per second. The OMAP3530 offers an increase of five times, enabling up to 10 million polygons per second. It also incorporates pixel and vertex shader functionality, API support and anti-aliasing for better image quality. As mobile games and applications such as GPS become more complex, this enables better control and performance speed. This core has been reduced in size from 6.5mm2 to 5.5mm2 due mainly to the lithography reduction.
The imaging and video accelerator (IVA) has been upgraded in the OMAP3530 to version 2.2. It contains a 430MHz TMS320C64x+ DSP core. This core has nearly doubled in size, going from 3.5mm2 to 6.1mm2, but the DSP core is considerably more powerful than the core in the previous OMAP2420.
The DSP has been improved from a TMS320C55x DSP at 220MHz to an advanced Very-Long-Instruction-Word (VLIW) TMS320C64x+ DSP Core. This has six advanced logic units, and each can support single 32-bit, dual 16-bit or quad 8-bit arithmetic per clock cycle, enabling multiple applications to operate simultaneously.
Other members of the OMAP3 product family extend the functionality even further. The OMAP3430, for instance, can display video at 720p resolutions -- 1,280 x 720 pixels, which qualifies as high definition. While a cell phone display may not be able to take advantage of this high quality on its small display, there are some options that make this worthwhile. The phone could be connected to a video projector to watch videos, and portable pico projectors entering the market can be connected directly to cell phones. Imagine taking a dozen movies on your cell phone when traveling and watching them in high definition on your hotel room wall. Almost makes a video iPod superfluous, doesn't it?
Gregory Quirk is product manager at TechOnline, a division of TechInsights.
ARMED AND READY: TI's second-gen OMAP2420 mobile application processor includes an ARM1136 core running at 330MHz (the digital block is the ARM11 core with graphics accelerator at bottom right; the IVA block is at top right). |
BETTER YET: The OMAP3530 integrates the 600MHz ARM Cortex-A8 core (at bottom left in photo, with 2D/3D graphics accelerator; IVA 2.2 at right). |