SDK Approach Does Not Work

In the Episode 411 of the The Internet of Things Podcast, Stacey interviewed Mohammed Ansari, the Senior Director Business Development, Qualcomm Aware Platform.

The Aware platform is designed to be a cloud-based IoT service that ties into Qualcomm’s chips to provide telemetry, device management and services such as precise positioning and optimizing the network connection based on the quality of local network options. Ansari explains why Qualcomm has chosen to build a cloud and why he thinks that customers will use it (even though chip firms have not historically had success launching software or service businesses). He also describes how two of Qualcomm’s prior acquisitions will fit within this cloud offering.

What was super striking to me was the statement "the SDK approach just does not work". And I could not agree more.

We have been saying / doing this for years at Silvair. Delivering a complete software package, guaranteed to work seamlessly and integrate into the entire stack: with mobile applications, the backend cloud and more. The traditional "SDK Approach" (which "just does not work") offered by most (if not all) chip vendors has been to offer a set of software libraries supporting their chips. The problem with that is... it really does not work. Or - should I say - does not work anymore.

The underlying problem here is the complexity of the embedded systems today. From our own turf I can say it has grown by an order of magnitude over the last couple of years. A simple mesh networking stack in 2013 could be 30kB. Today it is 300 to 1000kB, depending on technology and features included. Integrating all that with an application which runs on the same chip is a challenge. But making sure it all works, and performs well, is an investment barely imaginable. On the testing side alone, our internal numbers are: 13 thousand unique test scenarios, with 30 million test runs per year.

We have never had a problem offering our integrated software to companies who had never done software before. They could not do it in the past and it has been obvious to them they will not be able to do it today. But somehow challenging is trying to sell software to software companies. They will most often try their in-house development approach and most often today they will find the "SDK approach just does not work". Qualcomm has just acknowledged that, based on their interaction with their chip customers trying to do the SDK integration / testing / tuning on their own. We have been doing exactly this for years. With a great success. But often being challenged by the "SDK" crowd. It is great to have such confirmations on our early bet on how the embedded software industry would transform.