A month-plus back, I provided a preview of ARM’s upcoming Cortex-A15 (aka ‘Eagle’) core, whose first publicly announced licensee was Texas Instruments. The ‘A’ in its marketing moniker exposes its “highly integrated application processor, primarily for mobile devices” primary focus, as a successor today’s Cortex-A8 and Cortex-A9 devices. And in addition to its consumer electronics concentration, I suggested that ARM aspired to find homes for it in trendsetting server designs.

Cortex-A15 isn’t expected to appear until at least 2012, however. And plenty of you are ARM-interested but have unique application needs for which a consumer electronics (or, for that matter, server) -tailored SOC won’t suffice. Enter (ironically) Texas Instruments again, with the company’s latest Sitara ARM-based embedded processors and Integra ARM-plus-DSP multi-core hybrid devices. Fundamentally, the advancements in both product family cases are fueled by a lithography shrink to 40 nm.

With the Cortex-A8 based Sitara AM3892 and AM3894, this process transition enables a clock speed boost to up-to-1.5 GHz. Normally, to accomplish such a performance feat on the Cortex-A8 requires either an architecture license from ARM that enables fundamental re-plumbing (such as the license owned by Intel-now-Marvell, Infineon-now-Intel, or Microsoft for as-yet-unknown reasons) or dynamic logic or other exotic implementation techniques (the 1 GHz speed spec on Apple’s A4 CPU was, in fact, the tipoff to me that Intrinsity might have been involved in the design). TI’s mum on whether it was a secret Intrinsity licensee or leveraged similar technology (all it will fess up to is ‘various circuit optimizations’), but the company concurs that the 40 nm turn of the lithography crank was a key factor in the 1.5 GHz accomplishment.

A higher transistor budget for a given-sized sliver of silicon also opens the door to higher levels of integration versus predecessor ICs, and TI hasn’t disappointed in this regard. Both Sitara processors contain a robust embedded peripheral suite:

• Two-lane PCI Express Gen2
• Two USB 2.0 ports
• SATA 2.0 interfaces for up to two external drives
• Two 32-bit DDR2/DDR3 external memory interfaces at up to 1.6 GHz
• Two Gigabit Ethernet MACs
• A HDMI transmitter, and
• Two-channel video I/O

To the latter two bullets, both Sitara SoCs support up to two tethered displays at up to 1920×1280 pixel output resolutions. The AM3894 additionally embeds an Imagination Technologies SGX530 GPU core, running at up to 333 MHz, in order to hardware-accelerate 3D graphics. The XAM3894CYG costs $43.10 (1,000); pricing on the GPU-less AM3892 was not announced by the company.

TI’s new Integra processors further up the integration ante, embedding a TMS320C674x floating- and fixed-point DSP core, running at up to 1.5 GHz, alongside the ARM system processor core. The remainder of the peripheral mix matches that of their pin-compatible Sitara siblings; as with Sitara, Integra SoCs come in base (C6A8167, $46, 1,000) and integrated GPU-inclusive (C6A8168, $49, 1,000) flavors. A DDR2 version (TMDXEVM8168DDR2) of the evaluation module common to both Integra and Sitara is now available for $1,895. A DDR3 version with video capture capability is expected to begin shipping in Q1 2011. The free EZ SDKs, now available for download from TI’s website respectively for Integra and Sitara SoCs, include ARM and (for Integra) DSP source code along with (ditto) “hundreds of ready-to-use DSP kernels”. At the moment, the SDKs support Linux; Android and Microsoft Windows Embedded Compact 7 support is also planned for Q1 of next year.

Admittedly, TI’s four new processors aren’t revolutionary breakthroughs; as the company’s overview page makes clear:

• They’re not the first Cortex-A8-based embedded processors in TI’s product line (see slower Sitara prececessors)
• They’re not the first ARM-plus-DSP combos, either (see the OMAP-L1x ARM9-plus-DSP Integra predecessors)
• Nor are they even the first Cortex-A8-plus-DSP products (see DaVinci alternatives)

As such, they’re perhaps more accurately categorized as somewhat predictable evolutionary ICs. Yet, speaking of evolutionary, their performance increases versus precursors will be welcomed in ever-more-code-bloated and therefore speed-demanding applications. Their tailored peripheral mixes may be a better general-embedded fit than DaVinci for non-imaging-centric systems. And the performance-plus-integration combination may not only keep TI present in evolutions of existing customers’ designs, but also with Integra, enable the company to kick out a current two-chip CPU-plus-DSP combo or few from new customers’ sockets.