ISA Vision Summit 2009: Growing influence of embedded software on hardware world

This session on day 1 of the recently held ISA Vision Summit had a good mix of speakers. Moderated by Anil Gupta, managing director, ARM India, the speakers included V.R. Venkatesh, Senior Vice President, Product Engineering Services, Wipro Technologies, Kishor Patil, MD & CEO, KPIT Cummins, and Raju Pudota, MD, Denali Software. This is a slightly longer blog post, so bear with me, friends.

The pic here shows Wipro's Venkatesh making a point, watched by Raju Pudota, Amil Gupta and Kishor Patil.

Kicking off the panel discussion, ARM's Anil Gupta highlighted the strength of the Indian embedded software industry. As per IDC, embedded software accounts for 81 percent of the projected share of overall revenues in 2008, at $5.98 billion. This will go up to $7.29 billion, while still accounting for 81 percent of the projected share of overall revenues in 2009. The projected share of the overall workforce in this industry segment stands at 82 percent -- at 125,663 -- which will be maintained during 2009, even as this figure rises to 149,978! Quite impressive!!

Incidentally, a recruiter recently requested information on the workforce numbers in the Indian semiconductor industry. I hope this partly answers your question, friend.

Gupta further added that embedded design had now entered several sectors such as automotive, aerospace and defense, consumer and home products, household appliances, industrial controls, infrastructure and construction, medical electronics, transportation and traffic management, security and telecom. In short, a bright future for this segment ensured, especially in India.

Trends in embedded design include: more demand for features, embedded is driving complexity, and prices have been generally constant/going down. As a result, all of the innovation happening has been giving new experience to the consumers.

Wipro's V.R. Venkatesh cited the example of medical devices, which are adding functions via embedded software. He presented the case of an efficient infusion pump, which ensures that the five rights of medication safety -- right person, right dose, right medicine, right time, and right way -- are never violated!

Another example cited was of adding functions in mobile devices. Such mobile devices are making use more dual core chip solutions to run multimedia and MIPS intensive apps on a separate applications processor. They use open operating systems (OS) such as Symbian, Linux, etc., and also have built in sensors, such as motion sensors.

Consequently, usability is now becoming the focus, rather than pure user interface of the mobile. On the impact of software complexity, he said that OSs and middleware are now becoming more complex to enable quicker and easy to develop mobile applications, and also develop complex mobile application with the right API support. He also cited new advances in automotive telematics and navigation. These are implemented through complex software and demanding more hardware features.

Challenges in developing embedded software
However, increasing embedded software has also brought its own challenges. Today, the share of software is ~50 percent of the total cost of development.

Some of the challenges while developing embedded software include multiple regulations; split personality: display (local and remote), compute and communicate; UI; low-power design, application specific accelerators; wireless as de facto connectivity; integrated sensors and geospatiality for enhanced applications; built for untrusted environments (security, virtualization); and integration with service providers and enterprise systems.

Hardware and software in an embedded system are complimentary to each other. Software (middleware and applications) should be used as a 'Differentiator' to add more winning features to any new product, he added. There is a need for a platform approach for embedded software development to enable scaling of features and usage across applications. Finally, developers need to keep the cost vs. functions vs. efficiency tradeoff in mind.

Embedded systems landscape trends
KPIT Cummins' Kishor Patil touched upon the growing need of convergence for hardware and software. According to him, the key driving forces are:
* Low cost and high performance;
* Low power and green;
* Maximum storage and least area/cost;
* Development: Faster TAT (turnaround time);
* Mechanical centric => electronics centric; and
* High value and low cost.

Trends in hardware include silicon shrink at 0.7x, technology challenges at 45nm and below, and business challenges -- high volumes for amortizing high mask costs.

Commenting on the embedded systems landscape: market trends and implications, he cited these to be: electronics and applications emerging as distinctive factors; increased electronics in automobiles (~100 MCUs/ECUs per car); silicon shrinkage reaching its limit w.r.t. geometry and costs; and enhancing system performance with the same hardware.

Content growth has been quite notable in automotive electronics. According to Patil, in 2000, an average automobile had 1 million lines of code, 20 ECUs, electronics worth $400, and software constituted 2 percent of the cost of the car. By 2010, an average car will have 100 million lines of code, 50 ECUs, electronics worth $1,100 and software cost at 13 percent of the cost of a car. Of this, 50 percent will be infotainment and 30 percent will be power train.

Impact on stakeholders
So what is the impact on the stakeholders? For OEMs, tier 1s, and semicon companies, it brings new business opportunities, and application specific solutions -- common/configurable hardware differentiated by software.

It allows R&D to migrate from proprietary interfaces to open and standard-based interfaces. The impact on software developers includes use of heterogeneous processors, managing parallelism, as well as dealing with scalability, compatibility and re-usability.

Embedded software's growing importance
Denali's Raju Pudota focused on current trends, such as growth in UMPC (ultra mobile PCs) designs; multimedia and automotive. For hardware, it means higher integration, multiple embedded processors in one SoC, and multiple microcontrollers (MCUs) with independent functions. Most importantly, embedded software is needed to make all of this work!

He said that more software is required to run all of the IPs integrated on the chip. These can be procured from hardware IP vendors, or developed in-house or contracted to third party providers. Also, different processors require different skills and capabilities. Finally, integration and embedded OS level capabilities. Incidentally, embedded software has become a requirement on the semicon provider. However, third-party IP has been evolving slowly.

Semicon providers' activities are manifold. These involve developing software for in-house hardware components, sourcing software from hardware IP providers, integrating various software components, and also test software offered to the system integrator.

These growing activities present its own challenges, typically: quality of software provided by hardware IP vendors, high integration time, software verification, and increased investment in software capabilities -- and emergence of a new area of core competence.

What can the ecosystem do?
Given this scenario, the ecosystem has a major role to play. These include:

Ease of generation of hardware-aware software -- define methods to abstract design to enable auto-generation of device drivers; define methods to auto-generate device drivers; few companies investing in this area.

Define framework/platform to integrate software -- similar to on-chip interconnect; leverage mature general software development processes; and customize to specific requirements of embedded area. Finally, make software offerings open-source; leverage large independent developer community.

Ease testing of embedded software
There is also a need to ease the testing of embedded software. Some points to note: Leading semicon providers have home grown software integration and testing platforms; making use of traditional methods -- hardware-software co-simulation, simulation acceleration, emulation, and FPGA testing. However, no standard methodology is said to be evolving.

Many industry solutions currently exist for hardware-software integrated testing, such as CoWare, SystemC, Mirabilis, etc. Then, there's also simulation accelerators (parallel processing), and emulators (FPGA based).

Challenges include: huge investment in model development, high cost of ownership, the ability of third party IP provider to enable integration, and large turnaround time per test. A proper framework for the integration and testing of IPs and embedded software is the need of the hour.

Pudota added that while this is a tough challenge, it would improve time-to-market for complex SOCs, develop a third-party IP ecosystem, and enable the semicon provider to focus on core competencies.