Get email delivery of the Cadence blog featured here
A lot has been written about AI and machinelearningdeeplearning, but a friend asked me, what does that have to do with Cadence, anyway? And what do you mean when you say, IP? I realized that I haven’t explained it yet. So here goes!
Imagine that you have a system like a home security system, and it has a sensor, like a camera. That camera is then connected to a central processing unit to interpret the data sent from the sensor, and then from that unit, action is taken based on how it interprets the data—unlock the door or call the police.
Seems pretty simple, right?
Well, here’s the thing. If we’re sending raw data to the central processing unit, your system must have two capabilities: the ability to send the data quickly to the central processing unit (that is, bandwidth) and a robust enough computer to not only determine friend from foe (AI) but also decide how it wants to react. Both of those aspects of the system can be incredibly expensive—both in terms of cost, and in terms of processing power (and, therefore, power power). And if your sensor is a radar or lidar unit, the “cost”, especially in terms of bandwidth, is even more!
A model of the sensors required in an automobile. The Sensor Fusion unit here is the CPU.
There is a way to get around these “expenses”. What if, in the camera sensor itself, you could use AI processing to perform the identification (“this is a person”) and classification (“this is a friend”) first, so all you need to do is transmit the results to the CPU—so all the CPU has to do is decide how to act on the data (“unlock the door”)? This way, you don’t need as robust a connection between the sensor and the CPU, and the CPU becomes much more lightweight.
Now imagine that your system isn’t a simple security system, but an entire car that incorporates cameras, radar, lidar, and ultrasound sensors—and the name of the game in automotive technology is keeping the system lightweight, power-efficient, and processes as fast as a car accident. With multiple CPUs processing umpty-squillion bytes of data—from navigation to rear-view cameras, from driver distraction sensors to level three and even level four autonomy—designers must lighten the bandwidth and processing load that the CPUs must be able to handle.
This is a perfect example of what the Cadence Tensilica Vision family of DSPs is for. In the vision sensors themselves are Tensilica Vision digital signal processors (DSPs) that provide the image/vision processing ability to then transmit the simplified data to the CPU(s). The IP, or “intellectual property”, is the information in that processor, and it has been developed by Cadence. It’s a chunk of well-tested, well-coded processing power that is verified and will work according to its specifications, so the designers of the system doesn’t have to.
A list of all the types of AI and Vision IP developed by Cadence. In green are the Tensilica Vision family of digital signal processors, or DSPs.
Problems solved! Now you have a lightweight CPU (or, in this image, a graphics processing unit, or GPU) that can process interpreted data from the vision sensor.
Okay, so now let’s think about that connection that has to exist between and within the sensor and the CPU.
The sensor has to be able to “talk” to the CPU in a language that it understands. Whenever you have data that must go from one place to another in your chip (or board or system), there must be some standard way that that data is communicated (for more information on why protocols are important, see Why Home Renovation Should Be Easy). Using Ethernet is one way of doing this. Another way is to use a USB connection, or PCIe, or SerDes or MIPI or any of a number of different methods.
But no matter the protocol, you have to translate that data using IP. Imagine that this interface IP is a box into which you plug the data in on one end (using your selected protocol), and what comes out is the same data that can be understood by the receiver (the CPU, or another part of the board, or even a chip), using that protocol. This IP has been thoroughly tested and is guaranteed to work according to its specifications, and the specifications of the protocol.
Voilà, that big portion of work that has to be done by designers is all taken care of!
In yellow are the protocols that can be used for interconnects in a processor.
So now you have processing done in the sensor using Cadence Tensilica IP, and you have connectors within your chip, board, or system using Cadence interface solutions.
Now, suppose you don’t want to use the Cadence interface IP; suppose you want to use your own design. When you use your own IP, however, it still must conform to one of the protocols, which was developed by various standards organizations. But this is a huge job—specifications for standard interface protocols are often hundreds of pages long. Deciphering these specifications and accurately modeling the protocols is a huge development effort requiring deep technical knowledge. And it can be a very expensive mistake if you were to get it wrong!
So Cadence offers Verification IP (VIP—no, not a very important person, but it is important) to test whether your IP conforms to the protocols required by the standard organization that developed them.
In blue are the various places that Cadence’s VIP can be applied.
There is a good chance that devices you use every day were verified using Cadence VIP. More than 500 customers have trusted the Cadence VIP Catalog to verify thousands of designs spanning every type of electronic product. Without IP, the speed of technology wouldn’t go nearly as fast, because it would take so long to design the durn thing.
It’s hard to explain what exactly Cadence really does in the Crazy Wide World of Technology. Whenever I’m asked what we do, I say something like, “We make the tools used to design die, chips, boards, and systems. And when I say tools, I don’t mean screwdrivers and soldering irons. I mean hardware and software.” I generally leave out the IP bit, because it would take a 1000-word blog post to get there.
Historically, Cadence went from the computer-aided design (CAD, hence “Cādence”) world to help design chips; now we help design chips, boards, and systems. In addition, we make the IP that allows designers to perform acts of practical magic—like making security systems that can distinguish friend from foe, or produce a level-three autonomous car, or a make mobile phone that can identify that it is you that is looking at it, or vacuums that will clean your house, or smart speakers that can order laundry soap at the moment you need it, or video conferencing that takes notes in the meeting, or speakers that can adjust to various sound levels, or…
P.S. Thanks to Paula Jones for the guest blog post, 30 Years of Innovation – We’ve Come So Far!