Pop Quiz: What kind of part is found on the BOM for many smartphones, feature-phones, cameras, camcorders, portable GPS units, tablets, and other mobile devices? HINT: Its size is as small as 2.5mmX2.5mm, standby current is in the microwatts, and cost is in the pennies.
What? You guessed an FPGA? You must have cheated! Oh, or, looked at the title for the article, or read ahead, or maybe you already know about SiliconBlue.
We have mentioned several times that it is near impossible to compete head-to-head with dominant, entrenched FPGA companies like Xilinx and Altera. As a result, companies that are trying to make a go as startups in the FPGA space are wise to adopt flanking strategies where they avoid direct competition. SiliconBlue is wise.
There are good reasons you have probably never heard much about SiliconBlue, even though they are currently shipping millions of FPGAs designed into very high-volume applications like those listed above. Rather than trying to deploy the army of FAEs and idiot-proof design tools that are required to sell FPGAs to the masses, SiliconBlue decided to focus on a small number of high-value high-volume customers – those in the mobile device space.
FPGAs have never been very good at mobile applications. When one typically thinks of FPGAs, one pictures large, expensive chips with high power consumption. That’s thought to be the price of admission for the flexibility afforded by programmable fabric. However, a few companies like SiliconBlue have been proving that’s not the case.
Although SiliconBlue devices are technically FPGAs, they are designed with a completely different set of priorities than the devices with which most of us are familiar. Instead of storing the configuration data externally on a flash device, configuration is stored on chip using a non-volatile memory technology that can be fabricated with standard CMOS processes (derived from the memory currently offered by Kilopass.) This makes the devices effectively single-chip, stand-alone capable – eliminating the need for configuration storage and components on your board. Everything about these devices was designed for small size and low power, resulting in a completely different set of architectural trade-offs from traditional FPGAs.
Nonetheless, at the core, the fabric is still a series of four-input look-up tables with registers attached – clearly FPGA. These FPGAs, however, are so tiny that a five-digit number can be manufactured on a single 300mm wafer. If you do some quick math, a boat might cough up somewhere in the quarter-million parts range. That’s good for the price tag. Many devices are sold package-free for the smallest, lightest possible configuration required for space-thrifty mobile boards. The result is super-cheap, super-tiny FPGAs with near-zero standby current that can add a bunch of capability to a mobile device.
Why would mobile device developers want to use FPGAs? Well, for the same reasons most people do, actually. As you probably know, most smartphones use off-the-shelf applications processors from companies like Qualcomm. With standard chipsets, just about anybody can build a smartphone these days. The problem with that is that everybody’s smartphone ends up being the same – and that makes it hard to stand out in a crowded market.
Companies that want to make their products succeed have to differentiate them in hardware. That means some custom development is required. With the skyrocketing cost of ASIC design, only the very few largest companies (like Apple) can afford to do custom chips to differentiate their mobile devices. For the rest of the world, FPGAs could offer the opportunity to include those eye-catching features that make consumers pick your product over the field of contenders. Those FPGAs have to be tiny, cheap, and low-power, however. That’s where SiliconBlue comes in, along with a couple of other competitors – namely QuickLogic and Lattice Semiconductor. QuickLogic’s products are aimed at adding very specific capabilities to mobile devices – like better-quality video at lower backlight settings. Lattice is focused more on mid-range FPGA applications. That leaves SiliconBlue as the lone supplier of general-purpose FPGA technology to the high-volume mobile market.
On a visit to SiliconBlue’s Silicon Valley headquarters, one is greeted by a “trophy case” filled with devices that contain the company’s parts. There is a plethora of smartphones, point-and-shoot cameras, GPS units, camcorders, and even 3D camcorders. The company says their devices are typically used for functions like port expansion, video and imaging, sensor management, and memory management. Customers use the FPGA to extend the feature set and the longevity of their off-the-shelf applications processors. FPGAs are great for things like managing multiple displays and cameras with different resolutions, translating between incompatible standards, and adding additional interfaces not provided by the standard processor. On sensor management, FPGAs can offload the applications processor from what would otherwise be polling-related tasks – with big power savings.
Recently, SiliconBlue launched their second generation of devices – ICE40 – based on 40nm CMOS technology. That makes their tiny little devices even tinier than before, (and cheaper) with other bonuses like lower power, more performance, and more density… exactly what one would expect from a process node improvement. The new family ranges from 640 logic cells to 16K logic cells, with 25 to 222 IOs, and in packaged sizes ranging from 2.5mm2 up to a comparatively giant 20mm2. Many devices are sold as chip-scale ball grid arrays where the bonding balls are fabricated directly on the chip with no surrounding packaging. If you’re used to designing conventional FPGAs onto your board, get ready for the tiny 0.4mm pitch, but mobile PCB designers already live in that world, so they won’t have any heartburn from the tiny design rules. Static power for the smallest of the new devices is 15uW, and it is only 1500uW for the largest device. Embedded RAM ranges from 32K bits to 384K.
Since SiliconBlue is nurturing a small number of large customers, they have an easier task in supporting their tool suite. They use standard tools for FPGA design, with proprietary place-and-route and third-party tools like Synopsys Synplify for synthesis and other common functions. SiliconBlue supports customers that want to do their own FPGA design, helps people that want to do FPGA design with some assistance, and will do the whole project turnkey for companies that want FPGAs delivered like ASICs – ready to roll. Often, in short-schedule, high-volume mobile designs, companies put a premium on getting done and out the door quickly versus struggling with the learning curve of FPGA design, so they’re frequently happy to outsource the design work back to the FPGA vendor – where the expertise lies in optimizing the design for the target technology. SiliconBlue also has a rich portfolio of IP – particularly for the type of tasks commonly required in mobile designs.
If you consider the size and application space of these FPGAs, there really is no reason they’d ever come in direct competition with companies like Xilinx and Altera. The similarity really ends in the conceptual architecture of the fabric. However, the kinds of benefits SiliconBlue brings to mobile designers are a welcome improvement to the landscape – something that designers in other domains have been enjoying with FPGAs for a long time.
Image courtesy of SiliconBlue