posted by Bryon Moyer
We’ve spent a lot of time on sensor fusion here, and there are three players that routinely come up: Hillcrest Labs, Movea, and Sensor Platforms. Frankly, the first two have featured somewhat more prominently because they do more selling of “shrink-wrapped” product (for lack of a better word), while Sensor Platforms has tended to play closer to the vest, working with specific clients to integrate their algorithms. We did see them in QuickLogic’s FPGA sensor hub solution.
This week, they signed papers with Audience that will make them a part of Audience. If you’re not familiar with Audience (as I wasn’t), you might wonder what the heck is going on here. And if you’re a Sensor Platforms customer, you might wonder what this all means for you. Let’s address both of those.
First off, Audience makes chips that help your cell phone (or whatever) acquire the capacity for the “cocktail party” effect. In essence, they make chips that computationally implement what’s referred to as “Auditory Scene Analysis” (if you can handle the mixed metaphor of an auditory scene). This is the process by which our brains take all of the sounds we hear simultaneously – which are nothing but mixes of longitudinal air waves with complex harmonic content – and somehow figures out how to segregate the sounds, assigns them to sources, and, critically, lets us focus on one set over another even when both sets have inputs at the same frequencies.
This is the essence of the cocktail party effect, where dozens of conversations are going on simultaneously, all with voices in more or less the same range, and yet we can concentrate on one conversation without having it get jumbled up by the other conversations. Cell phones aren’t good at this, amplifying the voice as well as nearby traffic, that squalling brat, and that damn parakeet in the background.
You might think of the auditory scene as the set of other clues and cues that the brain uses – which these days would probably be called “context.” And if you’ve seen Sensor Platforms’ CTO Kevin Shaw speak, you know that they’re all about context. It’s the next big prize after simple sensor fusion. And apparently they’ve been working with Audience on algorithms. At which point Audience decided they liked the technology and ponied up some cash to own it.
So what does this mean for Sensor Platforms’ business and customers? In their conference call, Audience said that they’ll continue to service existing customers and contracts. Going forward, it sounds like they plan a dual strategy: continue selling the software algorithms while also working them into their chipsets.
There are certain businesses – military and medical were mentioned as examples – where they see the possibility of adding value, but they don’t plan to develop hardware. So they can service these markets with software.
For those markets they do serve with hardware, they see themselves having a power advantage because they have hardware accelerators for much of what they do – and it turns out that there was a big overlap between the accelerators they already have and those that would be needed to implement the Sensor Platforms algorithms.
It also sounds like they’re bought off on continuing with the Open Sensor Platform project that Sensor Platforms and ARM recently announced.
You can find out more in the official announcement.
posted by Bryon Moyer
We recently looked at Applied Materials’ solution to the challenges of lining small vias: using cobalt. But those are through-dielectric vias. What about through-silicon vias (TSVs)? After all, they can be a thousand times deeper than a standard via, so if a standard via is hard to cover, imagine how hard it must be for a TSV.
Of course, we’re talking a wider via, but AMAT says that standard physical vapor deposition (PVD) tools do an inadequate job of coating the TSVs when applying the barrier, for lots of the same reasons we discussed in the cobalt story.
Their solution to the TSV issue isn’t quite as radical as a new metal; it involves tightening up the angle of dispersion for the metals, providing better coverage. With better coverage, the barrier can also be made thinner, saving cost. A thinner layer is faster to deposit, improving throughput (and reducing cost).
(Image courtesy Applied Materials)
In addition, they’ve built a production-worthy chamber for use with titanium rather than the more typical “proven” tantalum. Titanium apparently being cheaper than tantalum. Both can be integrated with the copper seed.
You can read more about their Ventura PVD in their announcement.
posted by Bryon Moyer
IP used to refer to hardware designs that could be purchased off the shelf. Actually, at first they were designs that wouldn’t really work for any real application without a consulting contract to adapt them. But, over time, “shrink wrapped” became more viable. The idea was to save design time.
That idea still holds, but we’ve replaced one problem – design of individual blocks – with another: assembling all of the IP blocks into a complete system. And these IP blocks are more than your grampa’s simple fast Fourier transform; these are typically complete protocols that need to run a software stack.
Once assembled, the system will run the system software that’s being written for the SoC in parallel with the hardware design –software that’s separate from, and likely makes use of, the shrink-wrapped protocol libraries that may accompany the hardware IP.
So the full project development process involves hardware designers getting hardware running – first in prototypes, then in silicon. Meanwhile, software guys are coding away, using both virtual prototypes of the hardware and, eventually, the hardware prototypes that the hardware buys built.
In order to accommodate this more complex flow, Synopsys has announced their IP Initiative. It involves a more holistic view of how IP is integrated into SoCs, and the idea is to make the IP and accompanying elements work out of the box so no time is wasted on things that have already been completed – all of the effort can go into integration.
The image below shows the bigger picture of what they’re trying to accomplish. It includes both existing elements (like the hardware IP) and new elements being released as of the announcement, like the prototyping kits.
The IP prototyping kits are intended for hardware engineers, and they include a working reference design out-of-the-box on a HAPS board. IP licencees will have access to the accompanying IP RTL. Meanwhile, the IP software development kits include tools and virtual platform models of the IP that, again, work out-of-the-box.
The final bit, customized IP subsystems, gets to the challenges of putting all of these pieces together and coaxing them to work. Individual IP blocks work out of the box, but assembling them into an SoC isn’t trivial. Synopsys offers services to help create subsystems out of blocks.
You can read more about their offering in their announcement.