#pankaj chugh

[chughp]

Improving R&D Productivity in Semiconductor Industry

R&D constitutes 15-20% of the sales in semiconductor companies. Top 10 semiconductor companies spent in excess of $30b in R&D in 2015. Not only is that a significant cost, it is a critical source of competitive advantage.

 Complexity of R&D in semiconductor industry continues to increase:

 The technological changes are causing a huge escalation in the complexity of R&D projects. The complexity is being further amplified through need for more complex use cases being sought by customers, e.g for system-on-a- chip (SOC) devices that integrate processors, analog components, memory etc. on the same chip. R&D costs too are going up fast.  For example, according to Gartner, on average, IC design costs are have risen from $30 million for 28nm, to $80 million for 14nm, and will go up to $120 million for 10nm. R&D complexity escalation is causing, and is being caused by transforming business models in semiconductor industry. It is one of the major drivers of recent consolidation in semiconductor industry and emergence of horizontal business models involving reuse of IP blocks. And as the traditional vertical integrated models are being broken by fabless and fab-lite companies, the importance of competing on R&D has never been felt more.

 With the ever-escalating costs of R&D, to justify the investments, there is even more pressure on semiconductor companies to attain higher market share in the markets they focus on. Speed of product launch therefore has become even more important, which in turn is doubling the pressure on R&D organizations already reeling under technology driven complexity.

  Need for a robust approach to estimating R&D productivity:

 All this has put R&D productivity under spotlight. Semiconductor companies need to make sure that they have a robust mechanism for tracking their R&D productivity. The R&D productivity is however is not simple to estimate and track. Time and effort in R&D are not simple functions of transistor count. Extra time and effort can be justified by the type of circuit, density of circuit, the extent of reuse etc. These factors need to be normalized to find the actual causes of varying productivity. Most often such normalization would require certain degree of subjectivity. It is very useful nevertheless to be able to compare normalized productivity of the projects.  

  Identifying the right lever for productivity improvement:

Once the data has been normalized, it opens up opportunities to identifying the right lever(s) that can help improve productivity of individual projects, and R&D organizations overall. The are two broad categories of levers that can then be explored and applied to to improve the R&D productivity:

 Comparing variance between projects: By comparing variance between their own projects, semiconductor companies can help identify best practices being adopted by one project that can be applied to others. There are a number of differences that could be studied, for example,

·           Time and effort variance between various phases of the project, e.g. functional specifications, RTL design, physical design, testing, validation etc.  Furthermore, there are sub-steps e.g. design review etc, that can also be compared to identify improvement opportunities across projects

·           Number of sites used by design teams and configuration of the design teams can have a significant impact on the project productivity. For example, by increasing the number of sites, it becomes easier it is to scale teams and source talent, but it also adds to complexity in coordination. Comparing productivity of projects with different configurations can help identify the ones that best work for the company

·           Differences in non-standard/mandated tools, e.g project management tools used between different projects. Project teams teams may use different tools to collaborate with colleagues, track progress, manage code repositories, and plan validations etc. These tools can have significant impact on productivity. Differences in usage of tools across projects could unearth significant improvement opportunities

·           Tradeoffs applied by managers in their projects: There are a number of tradeoffs project teams make e.g., between reuse vs design from scratch, higher defect density vs extent of validation in their first release etc. These tradeoffs can vary widely by projects, and comparing projects can help identify a number of opportunities

  Identifying systemic opportunities for productivity improvement: These are opportunities that may exist across the organization, irrespective of the variances across projects. Companies could start by focusing on activities that consume most time, or most effort on average and identify the improvement opportunities. Furthermore, they also need to look at the activities that may have most downstream impact to zoom into focus areas for systemic productivity improvements. There are a number of organizational and process decisions that can help companies improve productivity of the R&D organization. For example:

·           Engineering skill level and competence of the organization

·           Choice of EDA tools used in the organization

·           Main design flows the company adopts in their R&D organization

·           The extent of management support and attention given to the R&D teams

·           The type and depth of relationship with the customers etc

   Need for improving R&D productivity in semiconductor companies has never been higher. A robust data driven approach to R&D productivity improvements is not just about saving costs. It can make a huge difference in companies’ ability to compete effectively in a competitive landscape that is intensely competitive, where technological advancement is unrelenting and customers are ever more demanding.

[chughp]

A side effect of IoT we must not ignore

We are rightfully excited about what Internet of Things holds for our future. It will usher in huge economic benefits, often estimated in the range of 10 trillion dollars. It will change lives of people in emerging countries, give us access to improved healthcare, make us more secure, and make our cities much smarter.  

There is however one thing that we may be ignoring, that needs to be addressed as embark on the journey of connecting the world. Estimates are that there may be as many as 2 trillion sensors getting installed every year by 2020 (22 billion in Automotive industry itself).  These sensors will come in different shapes, sizes and weights, for example, depending on whether it is active or passive. Let us say each sensor will weigh around 10g on average. This means we could be generating an extra 20 million metric ton of eWaste every year, doubling the amount we generate today. eWaste constitutes only 2% of the American thrash in landfills, but constitutes 70% of the toxic waste.  Only 12.5% of the eWaste is currently recycled in US every year. Due to the size and dispersal of sensors, our ability to recycle them may be even lower.

Ironically, one of the touted benefits of IoT is better waste management! Let us hope our collective ingenuity will also find a solution to this side effect of IoT!

[chughp]

Are we getting better at looking forward, but compromising our ability to look backwards?

Last year the GDELT project published a paper on how crunching through the big data of history can help us spot patterns and work out where the world is heading next. The GDELT database monitors news (broadcast, print, web) from across the world in over 100 languages and uses complex computer algorithms to codify what’s happening throughout the globe. Certainly our ability to project the future will improve, as we apply more sophisticated ArtificialIntelligence to draw such predictions about the future

Earlier this week however, Vint Cerf, one of pioneers of Internet, in his recent conversation at the American Association of Advancement of Science expressed a different concern.  He said that the humanity could be headed towards a digital black hole as the digital objects are becoming unreadable as the technology evolves.  He said that we be putting misplaced confidence in the longevity of digital information.  The information GDELT and similar such efforts are trying to preserve is tip of the iceberg. The volume of data that is getting generated continues to grow exponentially, much of which (currently) of no obvious value. Much of this data is being stored with little thought on how it will be retrieved by historians of the future generations. We are at the risk of losing a lot of this information, compromising the ability of future historians to look backwards.

A few years back, I was chatting with a friend and client in the Aerospace industry responsible for archiving engineering data for the airplanes they designed. I was intrigued to learn that the company kept all their designs on printed paper, and not in digital format. My friend told me that that the ever-evolving digital world meant there was no safe way to store this information in digitized manner.

Till the time we are able to create what Dr Cerf calls a ”digital vellum”, it may be a smart idea for my friend to maintain the printed documents for their airplane design, and for you and I to print the digital photographs taken from our latest digital cameras

[chughp]

Making modular product design key as Technology and Luxury converge

The smart device and wearable technology products are increasingly turning into luxury goods. Apple, by getting Angela Ahrendts from Burberry made its perspective on the industry very clear. And luxury goods companies are also keenly working on making their products smarter. Tag Heuer, for example is rumored to be launching a range of smart watches. Ralph Lauren unveiled the Polo Tech shirt that includes sensors knitted into the fabric to track movement and gauge performance. Jonathan Newhouse, global boss of Vogue, has coined a new word for the sector –‘Techno-Luxury’

 Although there are many learnings from luxury goods industry that could be applied to the new techno-luxury world, there are some unique characteristics too. In particular, there is a divergence between the economics of Technology and that of Luxury that companies need to worry about

The technology lifespan of hi-tech wearable is constrained by the Moore’s law. The products launched some time ago command much lower value than the ones launched recently. And the perceived value drops very fast with time

On the other hand, one of the most important perceived values for consumers of luxury products comes from longevity. Remember the famous slogans, “You never actually own a Patek Philippe, you merely look after it for the next generation.” Or the one from De Beers, “Diamonds are forever”

To manage this divergence of economics, unlike the traditional luxury product companies, Techno-Luxury makers will need to focus a lot more on modularity of their products. This would allow the products to be technically upgraded easily and selectively, whilst retaining the physical form and appearance.

Indeed much of the innovation is happening in this direction. Misfit has unveiled an activity tracker module that hides behind a separate Swarovski crystal. Google has come out with modular phone that has a strong aluminum and steel frame but can be fitted with new camera, speakers, batteries etc allowing the consumer to upgrade as the technology evolves. Likewise Montblanc in its recent launch, rather than creating a smartwatch, retained a mechanical watch with the smart device built into the leather “e-Strap.” This annexure to the watch can be bought as a separate module

Keeping modularity at the heart of the product design will be key for techno-luxury products - Designers would need to make this one of the key design objectives very early on in the design lifecycle, allowing them the ability to maximize the value they capture from the customers in the long term

[chughp]

Could it be smarter to make products a bit less smart?

Many product managers compete hard on cool features that have the potential to generate great buzz. These features are intended to create customer delight and make the product stand out. Of course, adding these features in the product is most often a good idea.

Sometimes however, decision to add 'customer delighters' is not as simple and linear. Noriaki Kano, an eminent professor at Tokyo University of Science created a model that classifies customer preferences.  Besides the delighters, customers also need other features, presence of which does not necessarily generate customer delight, but their absence can lead to a lot of customer dissatisfaction. 

Unfortunately many Product Managers do not fully comprehend these basic needs and end up competing in the wrong game. They keep adding delighter in an attempt to differentiate, at the expense of these basic needs. As product lifecycles have shrunk, and the pressure on profitability is immense, the risk of getting it wrong has never been higher.

Consider for instance, battery life of mobile devices such as fitness gadgets, smart watches and phones.  Analyst studies indicate that the batteries cost only 2% of the overall cost of building most such gadgets. However their importance in keeping customers from being dissatisfied cannot be undermined. Yet, often we have seen companies launching products with energy sapping cool features, or compact shapes at the expense of time between charges.

As technologies evolve, the compromises between basic features and delighters can be better managed. Perhaps it may be a better idea to hold off launching some of the cool features in the meantime? 

[chughp]

Are we ignoring the value from the third V of Big Data

Amongst the three Vs of Big Data, Volume, Velocity and Variety, many companies focus on deriving value by addressing the first two, and relatively less from the third. However one could argue that the most transformative solutions could come from tapping into the third V – the Variety of data.

  Often companies solve Big Data problems within the confines of a specific organization function, with the variety of data available with that function. For example, the online sales functions tend to address how they can give real-time and granular recommendations to the customers by improving the characterization of the products and the customers through online browsing and purchase patterns. This taps the volume and velocity of data well, and leverages the few varieties of data that are accessible to the function. Solving such problem is certainly critical, but it is also worth noting that most competitors are likely working on the same very problem, and the customers have already started to expect personalized recommendations i.e. this is already becoming table stakes!

  Today, not only has there been significant proliferation in the variety, but also the technological ability to use this variety is unprecedented. There are machine logs that come from manufacturing shop floors, data coming in variety of formats from suppliers, social data from customers, demand forecasts from resellers, weather projections, competitive information…. The list goes on. 

  Effectively tapping this variety is where the big opportunities could be. Could for example, the online sales team improve the recommendations further by looking at the data coming from shop floors that predicts a likely shortfall due to supplier quality issues. The company could steer the customer towards a different product or variant by giving a price discount.  Similarly, could the manufacturing operations function improve the yield on the shop floor by more closely monitoring the data coming from the field service engineers and customer complaints showing up on social media?

  These problems are certainly more complex to solve, mainly because they need for companies not to frame the right answers of the traditional questions, but to frame the questions themselves – They need to look at all data sources they could potentially tap into, and imagine how they could use them, and what value would that generate. Furthermore it requires them to resolve the very tricky issues that can hinder the access to data variety, especially data ownership rights, cross-department data access, and the technology infrastructure to allow the data access. Certainly more complex, but it is worthwhile thinking about all this before competitors start thinking about it!

[chughp]

HTML5 will not lead to commoditization of smartphones makers

The debate on the future of App stores and OS lock-ins is raging on as HTML5 picks steam. Since the strength of app stores has been key competitive factor for smartphones, many people have already written obituaries for smartphone makers. True, the potential for App stores to lock in customers will get dissipated as more and more development happens on HTML5. But that doesn’t mean smartphones will be relegated to commoditized gateways, very much like PCs were a few decades back.

  For starters, smartphones are likely to be used in a many different scenarios, which means diversity of the devices will likely be a lot more than PCs. This will allow many more opportunities to differentiate than we saw in PCs.  The diversity will go beyond the base features such as camera, screen resolution, processing speed, form factor etc. We are certainly seeing a continued stream of innovation – Samsung Galaxy Xcover can be washed,  Blackberry allows segregation of work and personal information, and with Nokia you can now print your own covers. The need for diversity is even bigger if you consider the accessories that will support the use cases in hospitals, schools, logistics etc.

  Smartphones are a lot more use-case dependent devices. In a way, HTML5 will lead to reduction in cost of development of these use-cases, and the utility and adoption of the smartphones will increase. It will therefore lead to wider adoption of smartphones. Just that the levers of competition will change from walled garden of applications to true differentiation of the hardware

[chughp]

Three cheers for Big Data on Cloud

This week MapR announced that it broke the MinuteSort record, where it sorted 15 billion 100 byte records in less than a minute. This calls for a celebration. We are generating data at an enormous pace. This will continue to grow exponentially, as more and more devices come online and get connected through the Internet of Things. The real problem will be making sense of the data at a reasonable entry and ongoing price.

The previous record was set by Microsoft’s specialized software using private machine clusters that needed 27,000 cores. This record was set up a cluster of 4206 cores, built from Google’s Compute Cluster that will soon be commercially available, and minor tweaking of MapR’s distribution of Open Source Hadoop distribution.

 Big Data processing on cloud is improving fast. Last year, two friends, after having built and run Facebook’s data service, have set up Qubole, a cloud data platform. All this investment and startup activity will hopefully allow us to unlock large number of practical use cases from the big data, and foster further innovation