Beyond phones: Why Qualcomm is betting big on machine learning, VR and 5G


Qualcomm has made some big bulletins this 12 months, introducing its first 5G modem, promising gigabit LTE speeds, and maximum not too long ago saying the business’s first 10nm processor in collaboration with Samsung. Customers are tough so much for from their telephones in this day and age, past simply extra chronic for apps and video games.

The fad in opposition to twin cameras calls for specialised ISP , whilst stand-alone and smartphone primarily based digital truth, which is being driven through Samsung’s Tools VR and Google’s Daydream, are requiring leading edge compromises to narrow down right into a cellular shape issue.

Over the last couple of years, those new calls for have modified the best way that Qualcomm is coming near processor design, and it sort of feels that the purpose is to permit the corporate to cater to extra than simply smartphones, as we’ve already noticed with drones and digital truth.

Whilst the Snapdragon 835 is going to be subsequent 12 months’s flagship design, Qualcomm additionally seems to be construction on its present applied sciences for decrease chronic IoT gadgets, cloud computing, and machine studying functions too. Right here’s what the corporate has been as much as.

Machine studying and heterogeneous compute

Whilst a lot of the controversy round machine and deep studying focuses round cloud compute answers, there are a rising selection of use instances that paintings absolute best on edge and cellular gadgets. This is the place traits in heterogeneous compute are turning into more and more vital, and Qualcomm has been making strides on this space because the creation of heterogeneous processing with its Snapdragon 810, as did different SoC builders that made use of ARM’s big.LITTLE generation.


Machine and Deep Finding out initiatives are more and more abruptly, however require new answers too. Supply: Bloomberg

Within the cellular area, we first in reality began speaking about heterogeneous compute with the disclosing of Qualcomm’s Snapdragon 820 and how the corporate deliberate to enhance the efficiency and power intake of symbol processing and different duties through working them on the most efficient core within the SoC.

We’re now not simply speaking about a lot unfold around the CPU and GPU right here, however Qualcomm has lengthy been the use of its Hexagon DSP and Spectra ISP devices to dump some duties too. The speculation is that through choosing the best element for the duty, efficiency is going up and chronic intake is going down.

This pattern is positive to be a key a part of Qualcomm’s technique going ahead, in particular when used at the side of machine studying to enhance the options to be had to customers. Examples of machine studying packages vary extensively relying on the , and this isn’t simply restricted to cellular merchandise.

The car marketplace, drones, and good properties are all primed to use machine studying to provide customers enhanced capability. This will vary all of the approach from object and voice detection proper as much as self reliant using automobiles. Actually, Qualcomm already has a devoted car Snapdragon 820 processor designed with machine studying and communique in thoughts, even though the core options are similar to the smartphone chip.



Different machine studying examples may come with making improvements to tool safety via facial or voice popularity, to taking an image and having the instrument robotically be sure that your members of the family are in focal point. Kind of handiest 1 % of smartphone packages are these days applying machine studying, however World Information Corp expects this quantity to develop to just about 50 % of apps within the subsequent two to three years.

See additionally:

Qualcomm Kryo and heterogeneous computing defined

September 3, 2015

In fact, it’s now not simply Qualcomm and OEMs which can be going to be operating on machine studying, 3rd birthday party builders are more likely to have quite a lot of excellent concepts themselves. To be able to facilitate more straightforward and optimised building on Snapdragon gadgets, Qualcomm introduced its Neural Processing Engine SDK previous within the 12 months, which these days helps Snapdragon 820 collection processors. The platform helps commonplace deep studying frameworks, together with Caffe and CudaConvNet.

There’s additionally a rising call for for twin digital camera generation, iris and facial scanning, and digital truth, which all require an expanding selection of complicated compute algorithms to be run on lately’s smartphones too. On the other hand, cellular is restricted through very strict chronic and thermal constraints, which brings its personal demanding situations on the subject of acting those extensive duties successfully. specializations and heterogeneous compete are the keys to overcoming those issues in cellular.



Be informed extra:

What is machine studying?

July 7, 2015

There are quite a lot of imaginable activity sorts with machine studying, a few of which run higher on CPU kind , others on the GPU, and some on devoted like a DSP. Many of those duties additionally wish to be finished in parallel, so spreading workloads throughout other cores is crucial to convey this sort of capability to client.

Ultimately, Qualcomm envisions much more devoted modules integrated within SoCs to a great deal enhance the power potency of the compute heavy duties, estimated to be anyplace within the area of 4x to 20x extra environment friendly.

We will be able to have to attend and see what form of specializations and duties are maximum commonplace sooner than a devoted items of silicon are deemed profitable. Within the period in-between, Qualcomm’s Hexagon DSP, Spectra ISP, and array of smaller sensor processing devices, which enhance the CPU and GPU that customers is also extra aware of, are permitting the corporate to provide optimized for builders having a look to become familiar with those new demanding situations.

We’ve noticed a an identical take with HiSilicon’s new Kirin 960, which moved ISP into the SoC in particular to maintain enhanced symbol processing.

Augmented and Digital truth


Qualcomm’s instance of a imaginable long term pair of augmented truth glasses.

Machine studying and heterogeneous computing isn’t simply destined for smartphones and automobiles although, it is additionally crucial a part of Qualcomm’s imaginative and prescient for digital truth merchandise too.

The big variety of sensors for imaginative and prescient and spacial consciousness, blended with tough three-D graphics and a way smaller chronic funds than PC primarily based equivalents, signifies that cellular AR and VR platforms need to be particularly chronic and efficiency environment friendly.

Right here’s only a small set of examples about how other processing necessities might be balanced throughout a heterogeneous processor.

  • CPU – apps, messaging, electronic mail, climate, and so forth
  • Sensor Processor – movement monitoring, gyroscope, temperature, and so forth
  • ISP – twin / three-D imaginative and prescient cameras, eye monitoring, iris detection
  • DSP three-D positional audio and binaural simulation, object detection, facial popularity, gesture detection, noise cancellation, speech popularity and studying
  • GPU Actual time graphics, machine studying, and person interface
  • Modem – 4G LTE, WiFi, and 5G add and obtain for cloud processing

Even if augmented and digital truth will be offering customers very other reviews, there’s numerous overlap in the case of and instrument necessities, in particular on the subject of sensor and graphics processing, and those are in reality simply an extension of lately’s smartphone applied sciences.

Digicam sensor counts in VR and AR headsets may succeed in 4, 8, or upper relying on the use case, and eye monitoring is more likely to be key to put in force vital applied sciences for GPU potency like foveated rendering. On the other hand, those form of applied sciences require further processing chronic and are steadily tied into machine studying algorithms, which all ties again into devoted to make this all run successfully in compact cellular shape components.


Now, it is imaginable to supply many of those options with their very own devoted elements. A picture processor for object popularity, a devoted DSP for audio, micro-controllers to care for sensors, and a separate CPU to tie the gadget in combination. Even if extremely versatile, this is an overly expensive and extra developer extensive than buying an answer that packs all of this into one chip.

Qualcomm has turn out to be more and more targeted on offering whole gadget answers in one chip in recent times, as may also be noticed through the mixing of ISP, DSP, and sensor applied sciences immediately into its Snapdragon collection. This additionally lets in Qualcomm and OEMs to optimise the to provide those form of options as successfully as imaginable, with tight integration between modules for upper height efficiency.

There is some possibility and trade-offs in predicting the kind of options that OEMs are going to wish, however Qualcomm is betting that builders are in search of fast-to-market somewhat than extremely customized answers, particularly for rising fields like digital and augmented truth.

See additionally:

Which telephones include a Qualcomm Snapdragon 820 processor?

6 days in the past

5G on the middle



Whilst we would possibly know Qualcomm absolute best for its Snapdragon vary of software processors, enhanced connectivity – in particular having a look in opposition to 5G – is shaping as much as be on the middle of many long term linked reviews. This is applicable now not just for upper answer video content material, however for streaming VR and AR reviews, sending knowledge for computation within the cloud, and even transmitting location and driving force help knowledge to automobiles out on the street.

Qualcomm’s not too long ago unveiled X50 5G modem targets to be offering up obtain speeds as much as 5 Gbps via strengthen for 8 x 100MHz band provider aggregation for enhanced bandwidth, up from the 4 x 20MHz CA noticed into lately’s main modems. The chip additionally helps 28GHz millimeter-wave applied sciences within the type of Verizon’s 5GTF and KT’s 5G-SIG, which might each develop into long term 5G requirements. It’s a state of the art answer that can most likely finally end up powering the primary 5G smartphones and capsules within the coming years.

See additionally:

Verizon posts its 5G spec: first U.S. provider to take action

July 12, 2016



5G isn’t with reference to offering ever quicker knowledge speeds to customers although, it’s additionally about connecting up hundreds of thousands of small, low chronic internet-of-things (IoT) gadgets throughout house and business markets.

Qualcomm is ready for this too, with its extremely low-power cell modems designed for a variety of IoT gadgets. Those can strengthen a variety of goods from good structures or home equipment that may switch average quantities of knowledge, proper right down to good business tracking that can be situated on the mobile edge and would possibly handiest wish to switch 10s of Kbps somewhat than 100s of Mbps.

Particularly for those IoT eventualities, Qualcomm has its Cat-NB1 compliant MDM9206 and MDM9207 modems already on the marketplace. The MDM9206 can closing for a number of years on simply AAA batteries.

Within the broader image, making an early play for 5G is going to present Qualcomm a head get started on the subject of now not handiest powering 5G smartphones, but additionally all kinds of linked merchandise.

Web of items

Whilst we’re on the topic of IoT, it’s value noting that it’s now not simply Qualcomm’s vary of Snapdragon processors which can be going to be powering this expected generation revolution. Qualcomm additionally provides builders a variety of WiFi, Bluetooth, and cellularly linked merchandise whole with an built-in microcontroller with more than a few processing functions. Those fall underneath the corporate’s CSR, FSM, IPQ, and different built-in answer levels.


The selection of internet-connected issues is ramping up exponentially and is expanding call for for neatly linked processing programs. Supply: digireach

Moreover, Qualcomm is additionally within the midst of obtaining built-in circuit producer NXP at a price of $47 billion. No small funding. As soon as this is finished, Qualcomm could have get admission to to a much wider vary of of built-in circuit applied sciences starting from transistors via to ARM microcontrollers suited to the car marketplace and a variety of different electronics packages.

This will likely surely assist the corporate make bigger on the greater than 1 billion IoT gadgets already on the marketplace that use Qualcomm chips. The corporate predicts that there might be as many as 25 billion gadgets linked to the information superhighway through 2020.



See additionally:

Qualcomm to procure NXP Semiconductors for $47 billion

4 weeks in the past

On this regard, and around the cellular and car sectors, Qualcomm is having a look to supply a number of built-in answers that can accelerate the improvement cycle. This may also be noticed via Qualcomm’s expanding selection of building forums, from its Snapdragon Flight Construction Package, via to its Snapdragon VR820 reference headset design. In fact, there’s a trade-off in the case of chip measurement, stricter thermal limits, and upper prices if builders and producers don’t finally end up taking advantage of the extra applied sciences packed into Qualcomm’s silicon.

Qualcomm is surely maintaining its chips on the chopping fringe of rising client and generation tendencies, however this is simply as a lot a possibility because it is an fulfillment. With IoT nonetheless now not rather successful over the mainstream and many purchasers nonetheless hesitant in regards to the prices and advantages for digital truth, to not point out failed AR initiatives like Google Glass, there’s the danger that more practical, extra specialised chips may achieve a bonus within the cellular area.

On the other hand, if Qualcomm is proper and AR, VR, IoT, and good car are the following big fields in client electronics, the corporate is rather a ways out in entrance when put next with different smartphone SoC producers.


Marshmallow Man, blog spiritual leader, has strived to make AppMarsh an independent and free blog from world monetary system. He and his followers are exiled by Google monster.