Blast radius – Francois Baldassari, CEO of Memfault
Francois Baldassari has been at the sharp end of the Internet of Things.
The French cofounder and CEO of US startup Memfault has horror stories of failed updates that ‘bricked’ devices, leaving them unusable. “I’ve been there and lived those updates,” he tells Nick Flaherty during Embedded World (EW2023) in Germany.
Those experiences, and the ‘blast radius’ of disruption around them, ultimately led to the development of Memfault’s platform for device reliability engineering for thousands or even millions of devices out in the IoT. “We think hardware faces a massive need for making software work better,” he said,
It is now branching out from its roots on ARM to move to RISC-V and other architectures.
Pebble
He was working on the embedded software team at US smartwatch pioneer Pebble. Every week, thousands of customers reached out to complain of Bluetooth connectivity issues, battery life regressions, and unexpected resets. Investigating these bugs was time-consuming and difficult. So they tapped into the ‘DevOps’ world and built crash reporting, cloud-based software delivery, and performance monitoring infrastructure so they could move fast, even when managing device fleets in the millions.
“The company is firmly rooted in the US,” said Baldassari (above, right). “I met Chris [Coleman, CTO] in college and we know each other very well. We were at Pebble with a complex device running on a STM32 developing fully sandboxed applications that you could download on to the watch. One of the things we struggled with was quality. As software becomes a bigger part of design more bugs escape. We had lots of those at Pebble and we built a way to automatically capture these problems.”
“It was very difficult but it worked. Eventually it ended up solving huge issues with our customers without them emailing us, identify the root cause without access to the device and look into a customer device.”
“We built it again at Fitbit and Occulus and so we started in 2019 and raised $8m in the Series A from a Paris investor – it comes down to the individuals – Jean [Sini, General Partner at Partech], the the partner who led the investment is a CTO and we got along well and he understood what we are doing.”
Investment
“Our R&D office is in Berlin and we have 20 people in Europe and we have a large presence in Boston and New York City. That led to the $24m Series B funding last year from a New York investor and it’s very clear what we want to do: over the air (OTA) updates, performance monitoring and device debug.”
“We think that these make up the basic iteration mechanism for DevOps but as we look ahead there are many more DevOps practices that we can bring from software to hardware, things like experimentation frameworks, AB testing, run time security, we already have an agent on the device that can monitor for security. Making sure the product is secure is something we did from day one. We work with medical, payment systems which have audited us every which way,” he said.
The Memfault software is available for Linux, Android, RTOS, and bare metal device developers and is compatible with a wide range of wired and wireless protocols such as Bluetooth, cellular, WiFi, LoRa, Thread or Zigbee (now Matter).
European chip partners
Security of the data is all senses is vital. The company has been building key relationships in Europe with companies such as Nordic Semiconductor, NXP and Infineon Technologies as well as Silicon Labs and Alif Semiconductors in the US.
“We are increasingly developing partnership with other vendors – we have a well developed API, to take our data and use it other places. Some customers use that for their own dashboards, our partners use it.”
“Ultimately the data belongs to the customer and they chose how it is used. If a customer wants to connect Memfault with another provider it is the customer that authorises the connection. We do not offer customer data to our partners.
That also applies to the hardware. “The industry continues to evolve and we want to make sure we are on the right platforms and we integrate into all the hardware that our customers care about.”
The software runs on ARM microcontrollers using the Common Microcontroller Software Interface Standard (CMSIS), the a vendor-independent abstraction layer for microcontrollers that are based on ARM Cortex architecture, but Memfault is adding new platforms.
“We released support for RISC-V last month, we think that there are other architectures such as AURIX, TrCore in automotive, even PowerPC,” he said. “We are enthusiastic users of CMSIS but we would love to work with the development tool vendors,” he said.
Long term IoT support
Long term investment is also key to give industrial customers confidence of long term IoT support.
“We’ve always made sure we had the financial resources to serve our customers for many years to company so we have long term roadmaps and our financial planning is long term and we come from a hardware mindset,” he said.
“In the same vein, it is very important for us to continue to serve our customers so we are not looking to sell into a company that will sunset our products, We think the hardware faces a massive need for making software work better so we think there’s a big opportunity to grow, we have multiplied our revenue every year but we will always make sure we can focus on serving industrial customers long term.”
Business model
That extends to the business model, which can be a challenge with millions of devices out in the field.
“We are a SaaS [software as a service] platform with a fixed fee based on the features that are enabled,” he said. “Then we have the data processing fee based on the concurrent number of devices that you expect to connect to the servers. If you have a million devices out there but they are intermittently connected with 50,000 that’s the data connection fee, that’s how we manage costs. For example you can have 1% sampling across a fleet, and we have a sampling engine that randomly samples devices.”
If a problem is found, the developers are notified remotely. They can then identify and fix the problem, and send an OTA update. However this can have problems, as Pebble showed.
“Our OTA platform never updates the whole fleet at once, so we have an update of 1%, then 5%, then 20% and reach stability milestones,” he said. “The reality is an update can be catastrophic and Memfault cannot magically stop that. So we limit the blast radius and use our analytics to notify team to identify and update. I’ve been there and lived those updates.”
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
