Arm TechCon 2019 Keynote Live Blog (Starts at 10am PT/17:00 UTC)

01:00PM EDT - Kicking off today is Arm's annual technical conference, Arm TechCon. Now in its 15th year, the company is looking to continue their long, successful run in the IP market, with Arm designs powering everything from toasters to servers. To do so, the company not only needs to keep current customers happy, a bigger challenge in the face of the success of ISA and IP rival RISC-V, but also to continue to grow Arm into new & expanding markets.

01:00PM EDT - For today's keynote, we've been told to expect specific announcements and discussions on IoT, autonomous vehicles, and what Arm is calling "Total Compute". So it sounds like Arm has an interesting show ahead.


01:01PM EDT - And here we go

01:01PM EDT - The bass is thumping. The room is shaking (or we're having an earthquake)

01:02PM EDT - Now on stage, Simon Segears, Arm's CEO

01:03PM EDT - Simon is quickly recapping the history of TechCon, and the history of Arm

01:03PM EDT - Smart phones happened, and the Arm landscape changed forever

01:04PM EDT - Arm is now on their fifth wave of computing

01:05PM EDT - Arm is now up to 150 billion chips shipped based on their IP

01:07PM EDT - One of Arm's current focuses is IoT. Arm wants to help its cusomters combine the physical and digital worlds by developing IoT devices to collect all sorts of data about the physical world

01:07PM EDT - And once you have that data, you probably want some AI to analyze it

01:08PM EDT - AI in the cloud, AI at the edge, AI in endpoint devices

01:10PM EDT - One example of IoT: a sensor to identify a blocked pipe

01:11PM EDT - Arm wants to move the data processing from the cloud to the local device. Slinging that data around is relatively expensive

01:12PM EDT - And after that: ioT devices running on 5G, skipping the need for a local network

01:14PM EDT - And with all of those devices in the world, Arm doesn't want security ignored either. Those devices must be secure - and Arm is developing technologies to do just that

01:15PM EDT - "There is no end to the ingenuity of bad people"

01:16PM EDT - One such technology being investigated by Arm is an architecture to isolate breaches

01:17PM EDT - Now shifting to discussing business matters. Simon is recapping Arm's flexible access program, which they announced this summer

01:18PM EDT - Flexible access allows extensive access to Arm's IP for a minimal upfront fee, and then sign a manufacturing license agreement once a company is ready to go into production

01:19PM EDT - Flexible access has been 75% of Arm architecture licenses signed in the last two years

01:19PM EDT - (CS grads represent! Today is Ada Lovelace day)

01:20PM EDT - Arm is launching a new program today to allow clients to add custom instructions to their Arm architecture chips

01:21PM EDT - This is for Armv8-M processors (so highly embedded parts)

01:22PM EDT - This program is focused on adding a handful of instructions; new instructions that would greatly benefit the sopecific workload a chip has been tasked with

01:23PM EDT - Think encryption and the like

01:23PM EDT - The first chip to get custom instruction features will be M33, which is coming in 2020

01:24PM EDT - And all future M-series CPU cores after that

01:24PM EDT - Custom instructions will be free to all customers

01:25PM EDT - I'll dig into this more later, but this sounds like a reasonable extension, but also a response to RISC-V, which directly competes with Cortex-M

01:25PM EDT - And that's Simon. Now on stage, SVP of Automotive and IoT, Dipti Vachani

01:26PM EDT - "What are the missing links taking us from prototypes to production?"

01:27PM EDT - Dipti (and Arm) want to get self-driving vehicles into the real world

01:27PM EDT - To get there, there needs to be proven safety. Standards and certification for safety

01:29PM EDT - As well as safety is proven software and compute infrastructure

01:30PM EDT - Arm offers a series of Automotive Enhanced (AE) processors, which are designed with a safety-first mantra

01:30PM EDT - Arm and its customers need to do security right. But they are also in a race to quickly get to market

01:32PM EDT - In order to get there on the software side of matters, Arm has been working with both commercial and open source vendors to support their software

01:33PM EDT - Arm has also been partnering with vendors like Swift Navigation, to develop a centimeter-accurate GNSS software stack

01:34PM EDT - Now for compute. ""How do we solve the compute problems" for automotive computing

01:35PM EDT - CPUs, GPUs, ISPs, and machine learning processors

01:35PM EDT - "Heterogeneous compute is critical and necessary" for autonomous compute

01:36PM EDT - And this heterogeneous compute stack needs to appropriately scale in performance

01:37PM EDT - Multi-core NPUs, meshes of NPUs, etc

01:37PM EDT - "The fourth missing link: collaboration"

01:37PM EDT - Arm has founded a new consortium, the Autonomous Vehicle Computing Consortium (AVCC)

01:38PM EDT - Members include chip makers like NVIDIA as well as automotive companies like General Motors

01:39PM EDT - Arm wants to bring together some of the "greatest minds" in the industry to help solve the problems in bringing autonomous driving to the market

01:40PM EDT - In recap: safety, software, compute, and collaberation

01:40PM EDT - Now on stage: Ian Smythe, VP of marketing for client products

01:41PM EDT - "The future for all mobile will move to digital immersion"

01:42PM EDT - "Digital immersion will engage all of our senses"

01:43PM EDT - And there are already multi-billion dollar industries built around digital immersion

01:44PM EDT - One challenge Arm wants to solve is getting video production and green screening entirely into a mobile device

01:46PM EDT - Total Compute: making cores besides CPU cores easily accessible by developers

01:48PM EDT - Arm will add a new instruction after Hercules, in Matterhorn, called MatMul

01:48PM EDT - MatMuil will double CPU GEMM performance

01:49PM EDT - Step 2 for digital immersion: security

01:50PM EDT - Mobile devices hold immense personal data. Therefore they need to be trustworthy

01:50PM EDT - Security will come in layers

01:50PM EDT - Platform, in-process, and application

01:51PM EDT - Matternhorn will introduce what Arm is calling "Secure-EL2"

01:52PM EDT - Isolating individual processes within secure memory to avoid cross-contamination

01:52PM EDT - There will also be stronger protections against return oriented programming exploits

01:53PM EDT - Finally, detecting memory safety vulnerabilities

01:53PM EDT - Step 3 for total compute: software & tools

01:53PM EDT - There are over 23 million software developers

01:54PM EDT - Arm wants Total Compute to be the platform of choice for all developers

01:55PM EDT - Arm is announcing today that Arm and Unity are extending their partnership, integrating Arm support into Unity

01:56PM EDT - Now recapping Unity and the importance of it

01:57PM EDT - The partnership will allow Unity to improve performance against the CPU, GPU, and NPU

02:00PM EDT - And that's a wrap on Total Compute

02:01PM EDT - Now on stage, Sha Rabii, VP for Facebook

02:02PM EDT - Dr. Rabii will be talking about augmented reality

02:03PM EDT - Recapping all of the big use cases for AR: work, reminders, HUDs, and more

02:04PM EDT - Even something simple like flashing a name will require an always-on contextual AI

02:05PM EDT - Not to mention all of thw compute required for the computer vision orocessing

02:05PM EDT - Facebook needs a dramatic improvement in silicon/efficiency

02:08PM EDT - Compute can be expensive. Data transfer is even more expensive

02:09PM EDT - Hitting their energy efficiency targets will also require intelligently picking optimal points on the voltage frequency curve

02:10PM EDT - Finding that point is a multi-variable problem

02:11PM EDT - Efficiency is also found in optimizing the AI workloads themselves

02:11PM EDT - High accuracy often isn't required, or only required at certain points

02:14PM EDT - Quantization-aware training remains a popular method to further reduce power requirements, by only using as much precision as is necessary

02:17PM EDT - (This is actually a really interesting silicon and algorithm discussion. Though difficult to keep up with it while typing on a laptop)

02:17PM EDT - "The next years will usher in a revolution in computing"

02:18PM EDT - And that's a wrap for the day 1 keynote. Thanks for joining us.