Lot of video stuff should be done in software on DSP: Sasken

BANGALORE, INDIA: Sasken, a communications focused company, sees communications as a large bar. Within that, there is significant place -- starting all the way from semiconductors up to the service providers Sasken is the company focused on delivering solutions in the R&D space across the entire value chain.

CIOL met up with Dr. G. Venkatesh chief technology and strategy officer and executive director of Sasken, to discuss a range of issues, such as the Indian semiconductor industry, the embedded space, DSP challenges, power management in devices, use of video IPs in SoCs, re-use of IPs, etc. Excerpts.

CIOL: How do you see the semiconductor space in India and what role is Sasken playing in this?

Dr G. Venkatesh: There are two ways to look at the semiconductor industry in India. One is, in terms of the kind of work that the Indian companies can do for the semiconductor majors worldwide in terms of design work, and in terms of work that goes around the chip design -- chips that they sell in order to create reference solutions for the end market places.

As an example, one can consider handsets as an end market. Then, the Indian companies are doing a lot of work today, basically in terms of developing chips and good handsets, and in terms of developing all the software that goes on top of those chips in order to realize the reference platforms that these semiconductor companies then sell to the handset makers. So that’s one direction.

The other way to look at it is India is an end market, which consumes a lot of electronic equipment. As India’s GDP continues to rise and then, we see that in the next decade, there is going to be a significant consumption of semiconductor chips in India. As India starts to consume the semiconductor chips, there will be local production, manufacturing, testing, support services and many other things. Design support services for this end market will also get more developed as we go forward.

CIOL: Sasken has been very strong in the embedded space too, and when we think of embedded we think of a whole range of applications. Can you enlighten us on what Sasken has been doing in this space and what’s the way forward?

GV: Sasken has been very active with semiconductor companies and system companies in developing chips. It basically means that you take a chip today, which is system-on-chip (SoC). If you have digital building blocks like analog building block, it will have bio components and it has a lot of software embedded on the chip. This software can be in the nature of drivers. It can be in terms of operating system blocks, it can be in terms of middleware blocks, it can be in terms of user interface blocks, it can be in terms of application fraternity and so on.

We help customers develop all of these components, including doing the digital block and the chip design itself. We help them in doing the embedded software so as to help them to validate the entire system before it is delivered.

CIOL: Since we are talking about building blocks, lot of DSPs do get into action. There are also lot of DSP implementation challenges involved. What are some of those challenges and how does Sasken tackle those?

GV: There are DSP cores, and developing the DSP core itself is a major challenge. However, we are not directly involved with that. I mean, we work with the customers like Texas Instruments and Analog Devices, who are specialists in developing DSP cores.

I think the real problem is to come up with an instruction set, which is good enough, on which you can program a large number of applications with good performance. At the same time, it must be low power.

Simultaneously, we should be able to develop an ecosystem of software developers who are familiar with this instruction sets and they are willing to develop the software for these kind of processes. So that’s the job of our customers, who are basically the silicon companies.

What we do is that we work very closely with these DSP core developers. While they are developing the instruction sets, we are figuring out how to map the various kinds of signal processing algorithms on to those DSP cores. For example, if you take a modem or a wireless modem, there are number of building blocks inside these modems, which require algorithmic programming, which in turn, requires you write those things in algorithms. Then you map these algorithms on to the instruction set of this process in an efficient way. That’s what we do.

Our specialty is in writing the algorithms and then converting those algorithms into fixed point cores, basically for the DSP process. We are able to do that in a way, which basically compacts the footprint -- which means the number of bytes that it takes to store this thing in the memory and the amount of run time it takes to execute a certain amount of data. We also look at things like how do we do it in a way that it reduces power consumption, for example.

CIOL: On power consumption, we all know power management is a very critical issue as far as devices are concerned. How do Sasken's designers handle power management issues?

GV: The hard part about power consumption really is that you have got to understand the entire system. For example, if you are looking at mobile phones, you might discover that a lot of power is actually spent in the display. Then you got to figure out when you need the display and when do not need the display.

Or, what are the kinds of things that you really need in the display, and then, knowing the application and knowing what is it that consumes a lot of power. Then, use the software to basically intelligently turn on and off things that are consuming power, without affecting the look and feel of the system. That’s where the intelligence comes in.

CIOL: There are also discussions centering around video IPs and integrating those into SoCs. How do Sasken's designers handle those?

GV: We are strongly present in video. In fact, a lot of our video IPs are shipping around the world. We strongly believe that except for certain acceleration blocks and so on, lot of these video stuff should be done in software on the DSP. In fact, that’s the path we have taken for the last decade.

We believe that a lot of videos can actually be done by programming a lot of the more advanced processes that we are seeing at present are in the embedded systems these days. We prefer to do everything simple.



As the video algorithms keep changing, we keep getting better and better algorithms. And putting it in hardware basically means that you do not have the flexibility to keep changing it. If it is in the software, you can replace one algorithm with another pretty easily.

For example, if you are selling MP4 and tomorrow find that H.264 is a better algorithm, then you need to use H.264. You can simply change the codec in the software and you are done. However, yes, there are certain elements -- wether you need to to do MP4 or H.264! There are certain elements, which can be accelerated and they can be in hardware as building blocks in the lower level modules. So, the video itself or the video processing itself -- we prefer to do it in software.

CIOL: Another aspect or point of discussion in the semiconductor industry is the reuse of IPs. How does Sasken look at this?

GV: A lot of these re-usable IP components come for silicon design. Actually, wherein one is looking at the semiconductor intellectual property portfolio that can be developed, building blocks can be used in multiple designs when you are doing the RTL.

As we are mostly focused on software, software is reused pretty much very well and understood. It is not that hard to do it in software as it is for silicon and that’s one of the reasons why we think we should put more stuff into software than into the silicon.