Bronco Webinar 800x100 1
Semiconductor Expert Forum

Posted on

Specialized AI Processor IP Design with HLS

Specialized AI Processor IP Design with HLS
by Alex Tan on 01-14-2019 at 12:00 pm
Categories: EDA, Siemens EDA

Intelligence as in the term artificial intelligence (AI) involves learning or training, depending on which perspective it is viewed from –and it has many nuances. As the basis of most deep learning methods, neural network based learning algorithms have gained usage traction, when it was shown that training with deep neural network (DNN) using a combination of unsupervised (pre-training) and subsequent supervised fine-tuning could yield good performance.

A key component to the emerging applications, AI driven computer vision (CV) has delivered a refined human-level visualization achieved through the application of algorithm such as DNN to convert digital image data to a representation understood by the compute engine –which is increasingly moving towards the network edge. Some of the mainstream CV applications are embedded in smart cameras, digital surveillance units and Adaptive Driver Assistance Systems (ADAS).

DNN has many variations and it has delivered remarkable performance for CV related tasks such as localization, classification and object recognition. Applying DNN data driven algorithm for image processing is computationally intensive and requires special high speed accelerators. It also involves performing convolutions. A technique frequently used in digital signal processing field, convolution is a mathematical way of combining two signals (the input signal and an impulse response of a system, containing information as to how an impulse decays in that system) to form a third signal, which is the output of these convolved signals. It reflects how the input signals impacted in that system.

The design target and its challenges
As a leading provider of high-performance video IP, Chips&Media™ developed and deployed various video Codec IPs for a wide range of standards and applications, including fully configurable image signal processing (ISP) and computational photography IP.

The company most recent product, a computer vision IP called c.WAVE100 is designated for real time object detection and processing of input video at 4K resolution and 30fps. Unlike the programmable software based IP approach, the team goal was to deliver a PPA-optimal hardwired IP with mostly fixed DNN (with limited runtime extensions). The underlying DNN based detection algorithm was comprised of MobilNets, Single Shot Detection (SSD) and its own proprietary optimization techniques.

The selection of MobileNets on top of an optimized accelerator architecture that employs depthwise separable convolutions is intended for a lightweight DNN. The four layer architecture consists of two layers (LX#0, LX#2) intended for conventional and depthwise convolution, and another pair (LX#1, LX#2) for pointwise convolution as shown in figure 2. On the other hand, SSD is an object detection technique using a single DNN and multi-scale feature maps.

As a DNN-based CV processing is inherently repetitious, evolving around the MAC unit –with massive data movement through the NN layers and FIFOs, the team objective was to have a tool that allows a rapid architectural exploration to yield an optimal design and shorten development time for time-to-market. The DNN based model was then trained using large dataset on TensorFlow™ deep learning frameworks. As illustrated in figure 3, the generated model was to be captured in C language format and synthesize into RTL.

In order to fairly assess the effectiveness of an HLS based solution versus the conventional RTL capture approach, two concurrent c.WAVE100 IP development tracks were assigned to two different teams. Such arrangement was done to mitigate risk of not disrupting the existing production approach which relies on manual Verilog coding captures. Furthermore, none of team members have prior exposures to the HLS tool or flow.

The team selected the Catapult® HLS Platform from Mentor as it provides algorithm designers a solution to generate high-quality RTL from C/C++ and/or SystemC descriptions that is targetable to ASIC, FPGA, and embedded FPGA solutions. A big plus on the feature side includes the platform ability to check the design for errors prior to simulation, its seamless and reusable testing environment, and its support to formal equivalence checking between the generated RTL and the original source. A power-optimized RTL, ready for simulation and synthesis can be rapidly generated through Catapult using the flow as shown in figure 4.

In addition to a shortened time to market at lower development cost, there are 3 key benefits pursued by the team:
– To enable a quick late-stage design changes at C/C++ algorithm level, regenerate the RTL code and retarget to a new technology.
– To facilitate what-if, hardware and architecture exploration for PPA without changing the source codes.
– To accelerate schedules by reducing both design and verification.

Flow comparison and results
At the end of the trials, the team made a comparison of the two flows as tabulated below:


The team takeaways from this concurrent development and evaluation efforts on design with HLS vs traditional RTL methods are as follows:

  • Easy to convert algorithmic C models to synthesizable C code. Unlike RTL, there was no need to write FSMs or to consider timing between registers. The C code was easier to read for team code reviews and the simulation time was orders of magnitude faster.
  • Optionally easy targeting on free software like gcc and gdb in order to quickly determine if the C code matched the generated RTL.
  • Ability to exercise many architectures with little effort using HLS, which otherwise was very difficult to do in the traditional RTL flow.
  • SCVerify is a great feature. There was no need to write a testbench for RTL simulation and the C testbenches were reusable.

To find more details on this project discussion check HERE.

Posted on

SOC security is not a job for general purpose CPUs

SOC security is not a job for general purpose CPUs
by admin on 01-14-2019 at 7:00 am
Categories: Automotive, IoT, IP, RISC-V, Security

Life is full of convenience-security tradeoffs. Sometimes these are explicit, where you get to make an active choice about how secure or insecure you want things to be. Other times we are unaware of the choices we are making, and how risky they are for the convenience provided. If you leave your bike unlocked, you can expect it to be stolen. However, we all know the feeling of learning that our credit card number has been stolen – clueless as to how or why usually. The other thing we need to be wary of is that hackers and bad actors are always looking for new ways to exploit security flaws. This means that things we saw as safe choices can, overnight, become risky.

Remember back in the day when you could easily use a debugger to find the code that did a password check and bypass it? Now we have protected address spaces and better encryption. System exploits are often found by hackers wearing white hats and then provided to vendors for fixing, before the public even hears about them.

However, in the last year a serious new security flaw known as Spectre has come to light that should give everyone pause. Like most people you bought a general-purpose computer with a CISC instruction set to both play games and do your banking. Processor vendors have spent the last several decades dramatically improving the performance of the general-purpose processors that are used in them. Among them are the Intel, AMD and sometimes ARM processors.

With the clock ceiling of ~4GHz for processors, CPU designers looked for ways to improve performance. An area ripe for optimization was wait states for memory reads, which can block processing for hundreds of CPU cycles. The widely adopted solution is predictive branching, where the CPU used prior execution profiles to determine the likely outcome of a branch decision. The processor would save state and proceed to execute the most likely code path. If the prediction was wrong the processor state was returned to the saved state. And, execution would resume with the correct branch. This seems safe enough…

Unfortunately, even if memory and processor registers are restored, there is still a latent trace from the code that was executed based on the prediction – the memory cache may have been changed based on memory reads. Downstream, hackers can use this in a number of ways to ferret out the contents of memory that was believed to be secure. One example is where the predictive branch was a memory bounds check, but the training led the processor to expect that the test would pass, but in reality, contains illegal memory accesses. The predictive code would then pull protected memory into the cache, where it can be retrieved later by additional hacker code. In fact, there are numerous other ways to exploit cache modification by malicious code leveraging predictive branch execution. Some even work in the Java runtime environment in browsers.

Unlike software exploits, this one relies on fundamental behavior of general purpose processors. So, when this exploit became public there was no fix ready, and in fact we will have to live with it, perhaps with some mitigation, for some time. The most secure fix is to disable predictive code execution, using the LFENCE instruction, but this leads to huge slowdowns in CPU performance. One security researcher estimated that 24 million LFENCE instructions would need to be added to the Office Suite.

Now look at all the new applications where processors are used that have heightened security requirements. In the face of this, it is time to start using different types of processors for different types of tasks – secure processors for critical jobs, and higher performance processors for less critical tasks. The push for heterogonous processors has been underway for some time, largely driven by performance needs. However, there is a growing need for specifically designed secure processor families. These might for instance be RISC based and are less vulnerable to predictive execution exploits. They also can have their own direct connected memory and security IP and accelerators, that are not accessible to any other part of the system.

Rambus outlines one such solution in their white paper “The CryptoManager Root of Trust”. Starting with a 32 bit RISC-V processor that is dedicated to security functions, the entire ensemble includes a number of essential components and the proper architecture for ensuring security. As such it is specifically designed to securely run sensitive code. It comes with dedicated SRAM and ROM memories. Also, there is an AES, secure SHA-2 hash core and asymmetric public key engine.

The Rambus CryptoManager Root of Trust (CMRT) also includes a true random number generator, a key derivation core (KDC) for deriving ephemeral keys from root keys. To detect tampering it also has a canary core that can detect glitching and overclocking. The Rambus white paper goes into detail about its comprehensive attack resistance. Also, it discusses the techniques it uses to create silos for sensitive code that needs to run securely. In fact, multiple roots of trust can be created to keep resources, keys and security assets for different application separate from each other. The CMRT core can be added as a complete security solution to SOCs to address the needs of a number of vertical markets. These include IoT, Automotive, Networking/Connectivity, and Sensors.

Rambus also describes the development tools and their provisioning infrastructure that complete the core’s development kit and deliverables. The white paper, which goes into much more detail on the full set of features and capabilities is available on the Rambus website. It is worth noting that the RISC-V core is not considered to be at risk from the Spectre exploit. I highly recommend reading the white paper, and its notes and references.

Posted on

CES 2019 The Year of De-Appification

CES 2019 The Year of De-Appification
by Roger C. Lanctot on 01-13-2019 at 7:00 am
Categories: Automotive, Events
2 Comments

CES 2018 saw the proliferation of digital assistant applications in cars (and homes, of course) with Harman International, Panasonic and Visteon showing multiple digital assistant implementations with in-dash infotainment systems. Panasonic showed a hybrid Alexa system capable of working off line and Harman showed a system with a dial to allow the user to select their preferred digital assistant: Alexa, Google Voice, Cortana or Bixby.

The leader in hybrid automotive speech recognition systems, Nuance, demonstrated a system capable of automatically selecting the appropriate digital assistant depending on the task. Meanwhile, German Autolabs was demonstrating an aftermarket device with multiple no-look, no-touch human machine interface options – including speech recognition and gesture – for communicating while driving.

For German Autolabs the over-arching message was clear: the age of de-appification had begun. Not everyone got the message in 2018, but CES 2019 is arriving in two weeks with an escalation in digital assistant integration.
No one is obsessing over recognizing accented voices or quibbling over when speech recognizers will be acceptable. Digital assistants have arrived and car makers and their suppliers are being forced to reckon with the consequences.

It’s not simply a question of accessing vehicle functions or cloud content or service resources. Voice interaction in the car is rapidly turning cars into browsers and driver and passengers queries into actionable and monetize-able inputs.

De-appification, an expression coined by German Autolabs CEO, Holger Weiss, refers to the reality that drivers and passengers will no longer be accessing content, applications and services via on-screen icons. The world of content and service delivery in the car will be an eyes-free and hands-free experience driven by voice.

More importantly, an increasing portion of the recognition and the process of responding to and/or anticipating driver and passenger needs will be supported by on-board artificial intelligence. The car will become more intelligent through the process of coordinating on-board sensor inputs, mobile device content and service information and cloud resources.
CloudMade was one of the first companies to demonstrate this capability. The competition to deliver this value proposition in 2019 and beyond will be fierce.

The entire vehicle will become an intelligent digital assistant anticipating driver needs and enhancing the driving experience. The most advanced systems will integrate with safety systems creating the opportunity for the vehicle to communicate and converse with the driver like the computer in “2001: A Space Odyssey” or like “Kit” in “Knight Rider.”
The implications for the development of in-vehicle systems in 2019 are significant and include:

  • The launch of OEM-branded systems such as “Hey BMW” and “Hey Mercedes” as front end interfaces to cloud partners including Alexa, Bixby, Google Voice, Cortana and Siri;
  • The integration of smarthome digital assistants – such as Orange’s new Djingo – with vehicle-based systems;
  • The capture of these driver (and passenger) requests to better anticipate driver needs and wants for integration with contextual marketing and payment platforms;
  • The demise of app-and-icon-centric in-vehicle user interfaces in favor of voice-and-gesture-centric systems. Like “Chris;”
  • The near elimination of human-centric call centers for concierge, roadside assistance and emergency services;
  • The enhancement of emergency response with artificial intelligence systems capable of instantly determining vehicle, driver and passenger status and automatically communicating the appropriate information to first responders and next of kin;

The enhancement of customer relationship management systems integrating with dealers to build stronger customer retention programs.

Multiple suppliers will use CES 2019 to demonstrate platforms designed to collect and interpret vehicle data. The next phase in this evolution, though, will be the integration with artificial intelligence and digital assistance systems intended to bring vehicles to life with a smarter, safer and more productive operating environment.

Will there by hiccoughs ahead? Definitely. Natural language systems capable of carrying on layered dialogues are still evolving and will take time to see market introduction. But it is not going too far to suggest that in-vehicle speech systems, by the end of 2019, will be capable of conversing with drivers to either help preserve alertness or to establish that a respite from driving is required.

In the end, what started out as a handy tool for the hands-free entry of destinations, the dialing of phone numbers or the selections of songs or radio stations, will begin saving lives with timely alerts and guidance. CES 2019 will usher in this new age of voice-based driver assistance.

Roger C. Lanctot is Director, Automotive Connected Mobility in the Global Automotive Practice at Strategy Analytics. Roger will keynote the Consumer Telematics Show on January 7 at Planet Hollywood. More details about Strategy Analytics can be found here:https://www.strategyanalytics.com/access-services/automotive#.VuGdXfkrKUk

Posted on

CES 2019 A New Era

CES 2019 A New Era
by Bill Jewell on 01-11-2019 at 12:00 pm
Categories: Events, Semiconductor Intelligence, Semiconductor Services

CES 2019 was held this week in Las Vegas and had over 4500 exhibiting companies and over 180,000 attendees. Over 6500 media and industry analysts attended (including yours truly of Semiconductor Intelligence). CES 2019 includes a broader industry than just electronics, which led to the show being renamed CES (previously the Consumer Electronics Show) and the sponsoring organization changing its name from the Consumer Electronics Association (CEA) to the Consumer Technology Association (CTA).

The CTA projected the overall U.S. consumer technology market will hit $398 billion in 2019, up 4% from 2018. Much of the total consists of the large dollar but slow growing categories of smartphones, laptop PCs, and televisions. These three categories total $131 billion in 2019 but are growing only 0% to 2%. These categories do contain some high growth products. 5G smartphones will emerge on the market later this year and are expected to account for only 1% of total smartphone units in 2019 but should account for 75% in 2022. 8K UHD televisions are also a new category in 2019 and should only account for about 200 thousand of 42 million TVs.

Most of the growth in consumer technology in 2019 is driven by emerging categories. Streaming services (video and music) are forecast to reach $26 billion in 2019, up 25% from 2018. This category reflects the broader CTA definition of consumer technology to include services as well as electronics. The major Internet of Things (IoT) categories are smart home (home controls, monitoring and security), smart speakers (such as the Amazon Echo and Google Home) and smartwatches. These three categories are projected to total $10.9 billion in 2019, up 15%. Another fast-growing category is in-vehicle technology including entertainment, navigation and driver-assist features. Totaling $17 billion in 2019, this category will grow 9%.

The emphasis on new consumer technologies was evident at CES Unveiled media event on Sunday, January 6. The event featured diverse applications such smart plumbing products from Moen and Kohler, numerous smart home products, an automated bread maker, a smart mirror to analyze your facial product needs and a pillow sleep aid.

Some of the products are of questionable practicality. Helite demonstrated an airbag for cyclists. The airbag resembles a life jacket and inflates when the bike crashes – protecting the torso. Any experienced cyclist knows the most common injuries in a crash are to the head (hopefully protected by a helmet), the arms and the legs. Ovis demonstrated a self-driving suitcase which will follow you through the airport. I guess pulling a wheeled suitcase by the handle is too much work.

An innovative product from French company E-Vone is a set of smart shoes which detect falls and notify caregivers with a precise location. Other solutions require the user to wear a device and push a button when the user needs assistance.

Flexfuel is another French company which develops products to reduce automotive pollution and increase fuel efficiency. Its location next to the bar at CES Unveiled gave it a new meaning.

The press conferences of the major consumer electronics companies focused on emerging markets. Panasonic emphasized artificial intelligence (AI), the internet of things (IoT) and robotics. The company demonstrated two products which use its electric powertrain platform: electric assist mountain bikes from Van Dessel and an electric motorcycle from Harley Davidson (available for pre-order at only $29,799).

Samsung’s CES press conference did feature its core businesses of smartphones and televisions. It has 5G networks up and running in South Korea and will introduce its first 5G smart phone later this year. Samsung displayed its 98-inch QLED 8K UHD TV, currently available for pre-order in the U.S. Samsung spent much of its press conference discussing its AI platform – Bixby – for televisions, cars and appliances. It also displayed a line of robots for health care (Bot Care), air monitoring and conditioning (Bot Air), retail assistance (Bot Retail) and mobility assistance (GEMS). Samsung’s emphasis on AI and robotics is evident from the layout of its huge booth (largest at CES 2019).

Sony’s press conference focused on tie-ins to its movie and music businesses. Its cameras, televisions and audio products were discussed in relation to these businesses. Singer Pharrell Williams made an appearance to discuss his visit to Sony’s R&D center in Japan. The only hardware product emphasized was Sony’s PlayStation 4 video game system.

Qualcomm began its press conference discussing 5G. The company expects over 30 5G devices (mostly smartphones) will be launched in 2019, almost all using Qualcomm’s RF devices. Most of the press conference was devoted to Qualcomm’s automotive products. The company said 30% of new cars are equipped with cellular connectivity and it expects this to grow to 75% in five years. Qualcomm’s booth was one of the largest at CES 2019 and featured 6 models of prototype 5G smartphones for China and Europe. Most of Qualcomm’s booth space was dedicated to smartwatches, noise cancelling headphones, and the smart connected car.

Intel’s CES booth was adjacent to Qualcomm and about two-thirds the size. It featured laptops using its latest Core i7 8[SUP]th[/SUP] generation processor. Like Qualcomm, Intel devoted most of its space to emerging technologies such as autonomous driving and immersive cinematic experiences.

What does the new emphasis on IoT, AI, robotics and smart cars mean to the semiconductor industry? It marks a new era. The major drivers of the past (smartphones, PCs and televisions) are showing flat to slow growth overall. Two of the largest semiconductor companies are shifting emphasis to the emerging markets. Intel, which gets most of its revenue from processors for PCs and servers, is pushing autonomous driving and entertainment delivered over 5G networks. Qualcomm, primarily a smartphone IC company, is also emphasizing automotive and IoT applications. The new era also opens up opportunities for other semiconductor companies to provide devices for the new generation of consumer products. Some of these emerging applications such as smart homes, smart speakers, autonomous cars and even personal robots may become as ubiquitous as PCs, TVs and smartphones over the next five to ten years.

Posted on

CES 2019 and Cycling

CES 2019 and Cycling
by Daniel Payne on 01-11-2019 at 7:00 am
Categories: Events, IoT
14 Comments

It’s January so time for my annual update on all things cycling that are being shown at CES this week. For 2018 my cycling goal was 11,440 miles, but an accident on September 1st cut into my goal, however I did reach 10,887.6 miles according to Strava.

eBikes
This category continues to grow in 2019, with many vendors offering us lots of choices from commuting to mountain bikes with electric motor assist.

Halfordshas embedded Alexa onto their smart computer bike at CES this year, which is a smart move because it isn’t so safe to use your fingers to click a touch screen while cycling, voice control is much safer. Pricing is about £1,000.

How about riding on gravel with an eBike? Van Dessel and Panasonic partnered to create the Passepartout Gravel E-Bike. Lots of my road biking buddies also own a gravel bike, but this is the first gravel e-bike that I’ve heard about.

Motorcycle vendor Harley Davidson showed off their concept of an eBike, although it looks more like my first minibike from the 1970’s:

Schaeffler has a four-wheeled, enclosed eBike called the Bio-Hybrid transportation system, and it is narrow enough to fit into a bike lane.

Ergosup and H2Tec Consortium have a hydrogen fuel cell powered eBike this year, now that has to be a technology first. Pictured is the fueling station, next to an e-Bike that runs on a hydrogen fuel cell.

In Portland, Oregon we’ve seen Lime land with their green colored electric scooters, but they will also be adding a commuter e-bike dubbed Lime-E.

Jeep is a rugged auto brand, so their e-Bike is also rugged and intended for mountain biking:

French company MOMODesign also has an e-bike similar looking to what Jeep has to offer:

JackRabbit is showing off an odd-looking e-bike with a smaller front wheel, aimed at city commuting.

French designer Jean Prouve came up with a retro looking e-bike, sold by Coleen:

How about a fat-tire, suspension, e-Bike? If that’s for you, then check out the MATE X e-bike on IndieGoGo.

Fitness Watches
Most serious cyclists opt for a dedicated computer, connected to their handlebars, however the smart watch companies will also track cycling activity. The Withings Move ECG smartwatch auto-detects cycling activity, monitors heart rate, and even has GPS for tracking. This watch tracks swimming, cycling and running activities.

Bike Safety
Cars have airbags, so how about using that technology for cyclists? Helite has a vest for cyclists that inflates as an air bag upon detecting an impact, saving you from getting bruises and broken bones. The technology uses a microprocessor, accelerometer and a gyroscope. I wonder if that device would’ve prevented me from getting a broken clavicle, two broken ribs and a punctured lung in my September crash.

Livall is a company from China showing off their smart bike helmet with lights and can send an emergency text through your phone if you crash.

You already own a bike helmet, but want to improve the visibility, so what kind of light can you add? Check out Rumble Lights on IndieGoGo:

Tired of carrying around a bike lock? How about a built-in bike lock on the front wheel? Bisecu has an app-controlled bike lock:

BIO-key shows off a biometric lock called the TouchLock.

From Cosmo City we have a smart helmet that auto-detects a crash, then sends out a Text or email message to alert your concerned contacts. I was on a group cycling ride on Saturday and another member wore a similar helmet with a built-in flashing light in back and front, very helpful to improve visibility.

AR meets cycling glasses in the Everysight Raptor AR Smart Glasses:

Indoor Trainers
Strava.com reports that in 2018 for the first time ever that their cycling members logged more rides on indoor trainers versus outside. On the high-end we have indoor trainers priced at $1,000.00 and up from brands like Wahoo, Tacx and CycleOps. My personal choice is the Wahoo KICKR while using the Zwift app along with the Discord app to chat with my buddies.

Posted on

WhisPro: A Speech Recognition Option from CEVA

WhisPro: A Speech Recognition Option from CEVA
by Bernard Murphy on 01-10-2019 at 7:00 am
Categories: AI, Ceva, Events, IoT, IP

In the superheated world of AI and Neural Nets (NN), many of us are familiar with object recognition in images: cars, pedestrians, cats and dogs and thousands of other applications. But there’s another class of applications, also growing rapidly, around audio AI. Early generations for command recognition in infotainment systems (eg navigation control) and smartphones (through eg Siri) were arguably more entertaining than compelling. In my view, what turbo-charged the audio AI market was Alexa. Now you can find products and active development in smartphones and personal assistants, watches, voice-activated TVs, appliances, headsets, hearables and head-mounted displays for VR/AR/MR. A FutureSource survey claims that growth in the audio market is now almost entirely driven by smart audio with a 15% CAGR from 2015 to 2018, accelerating to 30%+ from 2019 to 2023 (Yole Development).

When you think of an Alexa, Siri or similar, you might assume that all the smarts for these systems are in the cloud. That’s certainly the case for the more advanced functions (natural language recognition is one example), but it makes sense to keep some of the AI on the edge node, for all the usual reasons: latency, power or continuing to operate when a link is down. Consider always-on trigger-phrase recognition (“Alexa”, “OK Google”, …). It would be insane to go to the cloud to check every noise the device picks up, so trigger recognition has to be localized on the device. Or think about voice control for a home appliance such as a microwave or a thermostat where you don’t need to support an extensive vocabulary. Why go to the cloud at all? Why not handle all the recognition you need in the appliance (rather than “Sorry kids, cold leftovers tonight. The Internet’s down so I can’t turn on the microwave” :cool:).

CEVA has been active in the audio space for a long time, as you might expect for a company whose core expertise is built on DSPs. As smart assistants have taken off, they have become particularly strong in far-field voice pickup, the essential front-end of smart assistants and TVs for example. There’s a lot of technology here before you even get to neural nets because sound pickup isn’t quite as easy to deal with (in some ways) as light. First, it’s not as simple to figure out where a sound came from, which can be important if you want to separate the source from say TV sound. Figuring that out takes multiple microphones and beamforming. Then sound environments are generally a lot noisier that visual environments, not just thanks to multiple sources but also because sound echoes off surfaces; you have to filter out those secondary sources. A battery-operated device shouldn’t turn on even these stages unless it believes it has detected a human voice, so the first stage is a voice activity detector (VAD). CEVA has already introduced technologies to handle these aspects of voice pickup, in their ClearVox software running on various CEVA DSP platforms.

The next step is to take that clean, directionally-localized audio signal and run it into trigger phrase recognition – is this a command for me? At CES this week, CEVA announced availability of their WhisPro software which does precisely this. The software runs exclusively on a number of CEVA platforms and operates in always-on mode at ultra-low power. This is a big deal for battery operated devices. If you remember Amazon’s early introduction of battery-operated Echo devices, they worked fine but had to be recharged every few hours, even when minimally activated, because they were still burning power just listening for “Alexa”. Coupled solutions like WhisPro+CEVA DSPs are ultimately the only way to get to really low always-on power.

WhisPro is trained to recognize “Alexa” and “OK Google”; Moshe Sheier (VP Marketing at CEVA) tells me that it currently supports English and will soon support Mandarin and that CEVA will add training for other words/phrases at customer request. So a product could work equally well in multiple markets.

Important metrics in voice recognition are accuracy, noise resilience and privacy. You don’t want your smart assistant or TV turning on and asking you questions in the middle of the night. You also want to know that your commands won’t be misdirected en-route to the cloud. Moshe tells me that WhisPro, front-ended by ClearVox, delivers a recognition rate of 95%, comparable to Amazon and Google and is very noise resilient. Accept/Reject rates are also comparable to those providers. Also, these comparisons are at relatively short range; ClearVox extends the usable range to around 7 meters in noisy environments, which is likely better than other solutions can offer. And of course localizing initial recognition on the device naturally enhances privacy, just as it does in other AI at-the-edge applications.

WhisPro is available for licensing today for use on the CEVA-TeakLite-4, CEVA-X2 and CEVA-BX DSP platforms and is compliant with the main tier-1 voice services. You can learn more HERE.

Posted on

The Ups and Downs of Google Assistant Mini

The Ups and Downs of Google Assistant Mini
by Tom Simon on 01-09-2019 at 12:00 pm
Categories: AI, IoT
3 Comments

On Star Trek when they asked the computer to do something, they never heard it say “Sorry, you have no photon torpedoes connected to your account”. However, this sort of thing is something that happens at my house when I forget the exact name of a specific light. How did I get here?

I was reluctant to buy a “home assistant” for all the reasons – privacy, adding a complex device to my home, etc. Google, however, cleverly eased me into it. This is definitely a slippery slope, but I would not go back. It started on my android phone, with me asking it for directions and turning on the flashlight in the dark. Somewhere along the line I bought some Hue lights because I really like having lights that can change color and dim to suit the mood.

Then I bought a new house in Bend Oregon. I replaced probably 100 light bulbs – no exaggeration. Most were replaced with 2700K dimmable LED bulbs that use about 4 to 6W. The previous owners used only 65W incandescent bulbs with no dimmers. I added dimmers to most of them, but a few critical locations (bedroom, living room, deck, shower and tub, etc.) received Hue bulbs. At which point the minor annoyance of opening my phone to turn them on or control them, became a real nuisance. The phone app talks to the dedicated Hue hub.

The other contributing factor was my addition of a wifi controlled outlet for the water heater recirculation pump. This kind of pump typically runs all the time, keeping hot water flowing in a loop through the house plumbing. The alternative in bigger houses, is running (wasting) a lot of water each time hot water is needed. I reasoned that adding a remote-controlled switch would give me the best of both worlds, no wasted heat and pump electricity, and hot water for shower when I wanted it. Of course, for this a new app was needed on my phone.

Bottom line, I did not want to have to have my phone in my hand every time I wanted to turn a light on or off, or needed hot water. Enter the Google Assistant Mini. I’ll say, for the price it is an impressive bit of technology. First concern, it does a great job of hearing a command in a conversational voice – right away a lot better than Google Assistant on the phone.

The Mini is elegantly simple and easy to set up. Its one switch serves to mute the microphone. It has lights that indicate when it is actively listening and thinking. It is trained locally to recognize your voice, and it can use voice ID to limit access to your account. Account access is important when you start linking devices. In my case this includes the Nest Thermostats, Hue light hub, and attic fan and recirculation pump on Kasa wifi switches. It can also make phone calls and act as a speaker phone. And, of course it can play music through Spotify, but only on the Google device – it knows nothing of the Spotify running on your computer, Sonos, etc.

So, what are the catches? Yes, I like it and now have 3 around the house, but they are not without frustrations. When you add a Hue light it imports each of the 16 scenes for that light as a separate light. What’s worse is that it will not then use the scene names in commands. It is much better to just use color names. This is not just red, green and blue. Look up colors for paints or printing – hundreds – they almost all work. So, you can ask for fuscia, rose, sandalwood, etc. My suggestion is to just delete the scenes from Hue. Another thing is if you do not get the name of the light exactly right, Google disconcertingly says you have no devices linked to your account. Fear not, you just need to try again.

Another pro tip for the lights is if you have 4 lights, such as deck lights, just name them deck 1, 2,3,4, then you can just say ‘turn off the deck lights’, and they all get switched off, or you can address them individually. I see a complication coming in the future regarding light names. Right now, you can just say the name of the light, i.e. ‘OK, Google, turn on the bathroom’. The light will go on. But if the light is over a shower, you will say ‘OK Google turn on the shower’. Well this will be a problem when the shower faucet is also connected. I guess they will cross that bridge when they get to it, but it is unsettling at first.

The Nest integration is nice and works well. Google Mini makes it convenient to set timers, get the weather forecast, etc. We also have fun asking it silly questions, for which we occasionally get clever answers. Also, it is interesting to see how the system responds when your phone is within earshot. Clearly the Mini has precedence. They both will start recognition, but the Mini gets the upper hand. This makes sense, but definitely required some real-time cloud based processing.

I’ll follow up with another article on my experiences customizing commands. This is where things get interesting and where there will certainly need to be a lot of changes going forward. As a tech nerd, I was able to make it work, but the average consumer will be hopelessly lost I suspect.

Posted on

Samsung pre-announces miss on weak memory and phones

Samsung pre-announces miss on weak memory and phones
by Robert Maire on 01-09-2019 at 7:00 am
Categories: Semiconductor Advisors, Semiconductor Services
3 Comments

It should come as no big surprise that Samsung will miss its Q4 numbers. The company pre announced that profits will be 10.8T KWON (about $9.7B ) versus the 13.2T KWON analysts had predicted, close to a 20% miss. This number is also down about 39% sequentially. Revenue at 59T KWON instead of expected 62.8T KWON and down about 10%. The company blamed the miss on weak memory pricing and a more competitive smart phone environment.

This should not be a shocker to investors. The bigger shock is why, like Apple, analysts didn’t heed the many, many warning signs and cut number ahead of the quarter. The miss will likely drown out most of any positive tone coming out of CES this and dampen stocks that had started to recover from Apples bombshell.

TSMC warned about smart phones a long time ago and Memory pricing declines have been evident for a long time as well. Even if you don’t follow memory pricing on a daily basis, Micron talked about the weakened environment on their call.

Importantly for Samsung, the report clearly points out where their bread is buttered. Revenues only down 10% while earnings are off closer to 40% says that the memory chip business is highly profitable and likely one of the only things making any real money for Samsung. Its obvious that the vast majority of Samsung’s profits come from its chip business and the smart phone business is just a vehicle to sell more chips.

Its also clear that the report out on December 25th about Samsung reconsidering its chip investments will likely result in further reduced capex as putting any more memory chips on the street right now would completely crater pricing. If anything we might expect Samsung to slow production to match demand and hope that others, such as SK Hynix, Toshiba and Micron follow suit.

As fast as Samsungs memory spending went up …. it can come down even faster. We are also in Q1 now which aside from Chinese new year is also the weakest quarter of the year as it is a bit of a post partum depression from the holiday shopping binge (not) this season. This suggests to quick turn. At this point it seems that weakness, at least on the memory chip side , will last well into the year, if not to the end or beyond.

So much for the “one quarter air pocket”
We feel sorry for some analysts and investors who swallowed the “its only a one quarter air pocket” put out by a number of companies earlier in the year. Cyclical downturns, wether memory or logic driven, never last only one quarter as it takes a fair amount of time for excess capacity to be used up, idled or burnt off.

The news out of Samsung and Apple hits both memory and logic so the downturn is clearly both halves of the industry and as such will likely last longer.

If we assume the weakness started in the middle of 2018 (maybe earlier) its going to be minimally a year and realistically one and a half to two years at a minimum.

Could $50B in WFE capex become $40B?
Given the chart of capex spend in the semiconductor industry, its clear there was an unusual uptick beyond the normal inflection of spend. We could attribute a lot of this to 2D to 3D NAND conversion and some early EUV spend but a lot was just plain over capacity which will go away. We would not be totally surprised to see The $50B WFE capex shrink to a number closer to $40B…….

Companies will need to “reset” the bottom quarter
Lam had previously suggested that September was the “bottom” quarter. AMAT, smartly, has not yet called a bottom as its clear that its not. KLA talked about a “bounce back” in December then had to walk that comment back.

Given that Lam is all about Samsung and memory, unfortunately, Tim Archer may have to reset Lam’s “bottom quarter” statement on the December call as his first public outing as CEO. AMAT which is likely second in line of negative impact from Samsung probably just guides down a bit more. KLAC and ASML are more removed from Samsung memory but will have negative impact just the same.

The stocks
Its likely that we may have another down leg just as the tech stocks were starting to get over the Apple news. Its very hard to make a positive case for anything semiconductor related given that both memory and logic are down and the two largest consumers of chips, Apple and Samsung at about 10% each of the world chip market, are suffering.

It comes down to how long and how far down…..the stocks seem to keep hitting a resistance level which tells us we are near the bottom. The problem is that we could bounce along the bottom for quite some time.

If we wanted to stay involved with semiconductors (not that there is a good reason to do so…) we would look at names that are more logic oriented or outside of the core bleeding edge technology market.

In a perverse way. Intel’s lack of mobile exposure (with the obvious exception of modems) makes it less impacted than Samsung.

Who knows….if memory falls far enough maybe Intel could regain the mantle of the largest chip maker in the world. A title it lost to Samsung in the memory run up.

Also read:Apple as Apex of chip industry portends weaker 2019

Posted on

CES 2019 V2Vegas A Talking Car Turning Point

CES 2019 V2Vegas A Talking Car Turning Point
by Roger C. Lanctot on 01-08-2019 at 12:00 pm
Categories: Automotive, Events
3 Comments

The often pedantic debate over how cars will directly communicate with other cars and infrastructure will culminate suitably at CES 2019 with multiple C-V2X announcements and demonstrations. Foremost among those demonstrations will be parking lot drive-arounds at the Rio Resort Hotel put on by Ford Motor Company, Audi of America and Qualcomm.

C-V2X cellular technology takes advantage of a new air interface, called PC-5, which enables direct device communications at low enough latency to support crash avoidance applications for cars. More importantly, the direct interface works with or without network connectivity.

The participation of Audi in the demos at CES 2019 is notable given parent company Volkswagen’s determined commitment to dedicated short rage communication (DSRC) technology – instead of C-V2X – for vehicle-to-vehicle communications. Cellular-based C-V2X achieves the same outcome in the same wireless frequencies using existing LTE standards – and will also be available as part of 5G standards.

DSRC technology is already in limited deployment in some U.S. and European cities (China has notably taken a pass on DSRC in favor of C-V2X) such as Las Vegas for both roadside units and in traffic lights for communicating their signal phase and timing. The 35 Lyft-Aptiv “driverless” cars currently plying Las Vegas streets are already making use of DSRC technology to determine traffic light timings.

The Las Vegas roadside units were supplied by Commsignia, which will use CES 2019 to announce the introduction of dual mode DSRC/C-V2X modules. Like competing suppliers Auto-talks, Cohda Wireless and Savari, Commsignia has recognized the need to adopt and integrate the cellular alternative along with its DSRC solution in keeping with the spirit of Transportation Secretary Elaine Chao’s agnostic position regarding inter-vehicle communications.

Ford is waving the C-V2X flag particularly high announcing at CES 2019 its intention to deploy C-V2X technology across its entire vehicle product line by 2022. The announcement from Ford is as significant as Audi’s demo participation given that Ford was one of the founding members – with General Motors – of the Collision Avoidance Metrics Project (CAMP).

CAMP was founded nearly 20 years ago to develop the protocols necessary to support collision avoidance using 802.11p Wi-Fi technology. CAMP observers say the organization has recently come to recognize and accept that C-V2X is a formidable and acceptable alternative to DSRC – though there is no doubt that a lot of testing will have to be done.

By 2020, all Ford cars will be equipped with built-in cellular wireless connectivity from AT&T (in the U.S.; Vodafone in Europe; and China Unicom in China). Ford says the initial systems will enable over the air updating of the TCUs and vehicle infotainment systems with a broader strategy kicking in beyond 2020 by which time Ford will likely be deploying eUICC wireless technology allowing for re-provisioning of the wireless connection for use by different carriers in different regions.

The use of DSRC technology by the local Las Vegas Lyft-Aptiv self-driving cars is an indication that one of the most important applications for vehicle-to-vehicle and vehicle-to-infrastructure technology will be to better manage four-way intersections and to “see around corners” as some describe the technology. Regardless of which tech – DSRC or C-V2X – is used, CES 2019 will mark the mass market onset of V2X technology which might one day mean something to consumers who today are more or less scratching their heads. It’s worth noting that C-V2X technology will even offer the advantage of communicating directly with mobile phones – thereby enabling a pedestrian avoidance capability. With pedestrian fatalities on the rise in the U.S., that is very good news.

Posted on

CEVA-BX: A Hybrid DSP and Controller

CEVA-BX: A Hybrid DSP and Controller
by Bernard Murphy on 01-08-2019 at 7:00 am
Categories: AI, Ceva, Events, IoT

I’ve noticed hybrid solutions popping up recently (I’m reminded of NXP’s crossover MCU released in 2017). These are generally a fairly clear indicator that market needs are shifting; what once could be solved with an application processor or controller or DSP or whatever, now needs two (or more) of these. In performance/power/price-sensitive applications, like many applications in the IoT, market pressure often demands just one processor.


One way to cover this middle ground is to add functions to an existing architecture. So for example ARM provides hardware support for MAC (multiply-accumulate) operations, which you can use for example to build neural-net implementations on M-class subsystems. But it shouldn’t be a surprise that while an MCU with DSP extensions continues to be a good MCU and may compete on cost where DSP operation is undemanding, it is unlikely to be competitive in DSP-intensive applications. Conversely, a DSP amped-up to better perform controller functions is likely to suffer similar drawbacks in controller-intensive applications. In both cases, solutions can extend into the middle ground while not being ideal for the performance, power or value you would like to have.

Is this an artificial need in search of a problem? Not at all. The IoT has created a paradoxical requirement for very high levels of functionality (object and voice recognition for example) with responsive, even safety-critical performance and low power, while also being quite price-sensitive. Neural net workloads, modems, motor control and electrification in cars, bikes and scooters all need both DSP and control support with varying levels of demand between these two. You could require a DSP and a controller in every such application, a power and cost problem in many cases, or you could architect a true middle-ground solution. Which is what CEVA have done with their CEVA-BX product family, announced this week at CES.

Moshe Sheier (VP Marketing at CEVA) tells me that CEVA-BX is a new architecture, not an extension to an existing architecture, designed for the high-complexity signal handling performance and low power expected of DSP kernels, together with the high-level programming and compact code sizes expected for the established control code base. It supports parallel scalar compute engines and clocks at 2 GHz in TSMC 7nm, offering 4X the horsepower of the current CEVA-X2 DSP. The Instruction Set Architecture (ISA) supports Single Instruction Multiple Data (SIMD) widely used in neural network inference, noise reduction and echo cancellation, as well as native complex math and a range of floating-point options, in support of applications like high accuracy sensor fusion and GNSS positioning algorithms.

To meet control performance expectations, the CEVA-BX architecture incorporates features including a large orthogonal general-purpose register set for compiler efficiency, new Branch Target Buffer (BTB) for minimizing branch overhead, a hardware loop buffer for reduced power consumption of code loops, a fully cached memory subsystem, and native support for all standard C types. The CoreMark/MHz score of 4.5 places it comfortably in the MCU pack. CEVA-Xtend is available to add proprietary ISA instructions to the architecture to accelerate proprietary algorithms and to leverage CEVA’s automatic Queue and Buffer management mechanisms when integrating co-processors (for example in 5G PHY control) or creating a cluster of CEVA-BX cores.

CEVA-BX comes with extensive software libraries, both fundamental (DSP neural net, neural net frameworks, compiler, debugger and RTOS) and application (cellular IoT and GNSS ISA extensions, CEVA ClearVox noise reduction for voice-activated systems and CEVA voice recognition). Quite a starting package for application developers.

The family currently contains two members: CEVA-BX1 and CEVA-BX2 which differ in range of MACs offered. CEVA-BX2 comes with quad 32×32-bit MACs and octal 16×16-bit MACs and is intended to address very intensive DSP workloads like 5G PHY control, beam-forming and neural net workloads at up to 16 GMACs/second. CEVA-BX1 is more for the value to mid-range applications in narrow-band IoT communication and always-on (ie low power) voice-pickup or sensor fusion at up to 8 GMACs/second. These cores are currently available to CEVA lead customers and will become available for general use at the end of Q1 2019

Back to the market question. IoT devices are becoming a lot smarter than the dumb nodes we originally expected, not just in being able to do local processing but also in adding local intelligence so that communication with the cloud can be minimized and temporary loss of communication can be tolerated. That demands higher DSP workloads in communication, in neural nets, in motor control and other functions for which the traditional DSP+controller architecture is apparently struggling to keep up. The hybrid architecture approach is worth a look, especially since there is a lot of detail I skipped in this brief review. CEVA is presenting at CES this week so see them there if you can, otherwise check the product details HERE.