Custom, but without the Custom

  • Page 1 of 1
    Bookmark and Share

Custom. It means a design that’s exactly what it needs to be in order to meet an application’s requirements. It also means a design that requires a lot of non-recurring engineering (NRE) expenses for research, development, testing and certification. If only we can get the former (purpose-built) without as much of the latter (price tag).

Consider the medical device market. Medical imaging applications continue to attract custom R&D investments. Radiography, ultrasound, computed tomography and MRI systems produce detailed graphics to display on ever larger monitors. Patients and doctors need not pay attention to the rocket science “behind the curtain.” The rocket science involves all disciplines from design tools to ASICs to board-level designs to algorithms implemented in firmware (FPGAs and microcontrollers) or DSPs or even GPUs running OpenCL code. Capturing waveforms, signal processing and calculating 2D and 3D graphical maps are distinct areas that come together to round out a full custom system design.

It goes without saying that custom engineering is expensive. In a competitive marketplace, time-to-market and time-to-revenue are still everything, and project managers and executives alike scour the planet for special resources and building blocks to get a leg up. As medical device manufacturers continue to grow by acquiring the latest start-up companies, getting through FDA certification quickly means either the survival or demise of start-ups and their breakthrough technologies.

Advancements in embedded processors and ecosystems, along with FPGAs that have SERDES PHYs to hang on PCIe lanes, are accelerating the move toward modular architectures. “Modular” means comprised of individually upgradable building blocks. It opens the door for commercial sourcing of items like compute/control processor modules that run popular operating systems. For line-powered medical devices, the big computer-on-module winner so far is COM Express. Predominately x86-oriented, this standard from PICMG is adding value within the custom medical electronics community—for design and for operations/logistics alike.

Using an off-the-shelf processor module allows OEMs to get to market quickly without re-inventing the wheel by putting processor, chipset, RAM, LAN, etc. down on the main board. Instead, the main board becomes a custom carrier board with an off-the-shelf CPU subsystem plugged into it. COM vendors are quick to tout the benefits of modules, including future upgradability and multiple price/performance points with a common carrier board; a broader range than is possible with socketed processors alone.

There are two additional benefits of modules that aren’t mentioned often, yet are “priceless” to medical device manufacturers. The first is still within the realm of upgradability, but not after 5-7 years in production when the processor goes end-of-life (EOL). Rather, after 2-4 years of R&D and while still in product development, it’s imperative to upgrade to the latest processor technology just prior to submitting FDA paperwork and clinical trials and production. Sometimes this move is to squeeze a few hundred MHz more out of the processor. Sometimes it’s to reduce the thermal design power (TDP) for a given clock speed. A drop of 20-40 percent of processor power consumption can simplify the thermal design (cooling solution), or simply add a greater margin before the processor throttles down its speed. But the most common reason to do this is to re-start the 7-year lifecycle clock of processor availability. Medical system OEMs want to be in production for 10 years. At least, they want the ability to stay in production without having to tie up cash in stockpiling EOL inventory, and/or having to tie up engineering resources to re-qualify the latest processor module sooner instead of later. Processors and chipsets don’t come in pin-compatible replacements after 2-4 years, so using a module can prevent a board re-spin and debug cycle.

The second benefit of modules that is rarely mentioned has to do with the high cost of the carrier board components for imaging, whether DSPs or FPGAs. In production, any boards that don’t pass functional test need to be debugged or scrapped if not fixable. If the processor and chipset are also on the same board, the failure / scrap rate and cost go up significantly. It’s better to keep them on separate boards.

COM Express and other processor module standards make it possible to get the benefits—and then some—of a full custom embedded design without the full custom sticker shock. The latest quad-core processors with excellent integrated graphics lend themselves to the performance and long-term availability needs of the medical imaging market.