Smart Atoms Blog
Engineering insights from building, shipping and running connected products.
8,547 exposed control systems: secure energy devices start in the design
A study presented in October 2026 found 8,547 control systems at European solar and wind parks reachable from the internet. Operators can’t always secure the site network; manufacturers can make it matter less.
Intersolar Europe 2026: our display portfolio and the Energy Gateway accelerator
At Intersolar Europe 2026 in Munich we showed the LaMetric display portfolio and the Energy Gateway accelerator, and talked with energy device makers about getting connected products to market faster.
8,547 exposed control systems: secure energy devices start in the design
A study presented in October 2026 found 8,547 control systems at European solar and wind parks reachable from the internet. Operators can’t always secure the site network; manufacturers can make it matter less.
8,547 exposed control systems: secure energy devices start in the design
A study presented in October 2026 found 8,547 control systems at European solar and wind parks reachable from the internet. Operators can’t always secure the site network; manufacturers can make it matter less.
What stops a test build from reaching your customers?
In September 2026 a test firmware build reached customers’ smart fridges and some stopped cooling. Every team ships bugs; what matters is what stands between a build and the fleet.
The RAM shortage: two levers embedded teams still control
LPDDR4 prices rose sevenfold in a year and Raspberry Pi raised prices again on 1 October. We can’t design around a market-wide rise, but we can decide how hard it hits a product.
MIPI CSI-2 or USB? Choose the camera interface for the product, not the prototype
MIPI CSI-2 or USB for an embedded camera? If you are building a product, choose MIPI. If you are validating an idea, choose USB. The trouble starts when the prototype camera is assumed to be the production camera.
QR codes and AprilTags look similar. They solve different problems
QR codes carry information. AprilTags are built for machine vision: position, orientation and pose. On LaMetric Solar Glance we used AprilTags to measure a TV with a phone camera, with no AI required.
You upgraded the NPU from 2 to 10 TOPS. Why is your edge AI still slow?
TOPS is not edge AI performance. The NPU is one stage in a pipeline from camera to application, and the bottleneck may be memory bandwidth, CPU fallback, preprocessing, buffer copies or post-processing.
The most expensive mistake in edge AI: optimising the AI before the camera
Teams spend weeks on models, quantisation and benchmarks, while some of the biggest accuracy gains may come from the imaging pipeline. No AI model can recover information the camera never captured.
The first solution is usually more RAM. It shouldn’t always be
When LaMetric SKY outgrew 128 MB of RAM, the obvious fix would have raised the BOM cost of every device. Profiling, software optimisation and zRAM kept the original hardware.
Every embedded product starts with an SD card. Most end with eMMC
Embedded Linux prototypes boot from SD cards; most production devices move to onboard eMMC. What differs, when SD is still fine, common failure modes, and a checklist and test plan before mass production.
Intersolar Europe 2026: our display portfolio and the Energy Gateway accelerator
At Intersolar Europe 2026 in Munich we showed the LaMetric display portfolio and the Energy Gateway accelerator, and talked with energy device makers about getting connected products to market faster.
Engineering insights in your inbox
Plus new articles, news and company updates.
