Connect with us

Tech

Semron wants to replace chip transistors with ‘memcapacitors’

Published

on


A new Germany-based startup, Semron, is developing what it describes as “3D-scaled” chips to run AI models locally on smartphones, earbuds, VR headsets and other mobile devices

Co-created by Kai-Uwe Demasius and Aron Kirschen, engineering graduates from the Dresden University of Technology, Semron’s chips use electrical fields to perform calculations instead of electrical currents — the medium of conventional processors. This enables the chips to achieve higher energy efficiency while keeping the fabrication costs to produce them down, Kirschen claims.

“Due to an expected shortage in AI compute resources, many companies with a business model that rely on access to such capabilities risk their existence — for example, large startups that train their own models,” Kirschen told TechCrunch in an email interview. “The unique features of our technology will enable us to hit the price point of today’s chips for consumer electronics devices even though our chips are capable of running advanced AI, which others are not.”

Semron’s chips — for which Demasius and Kirschen filed an initial patent in 2016, four years before they founded Semron — tap a somewhat unusual component known as a “memcapacitor,” or a capacitor with memory, to run computations. The majority of computer chips are made of transistors, which unlike capacitors can’t store energy; they merely act like “on/off” switches, either letting an electric current through or stopping it.

Semron’s memcapacitors, made out of conventional semiconductor materials, work by exploiting a principle known in chemistry as charge shielding. The memcapacitors control an electric field between a top electrode and bottom electrode via a “shielding layer.” The shielding layer, in turn, is controlled by the chip’s memory, which can store the different “weights” of an AI model. (Weights essentially act like knobs in a model, manipulating and fine-tuning its performance as it trains on and processes data.)

The electric field approach minimizes the movement of electrons at the chip level, reducing energy usage — and heat. Semron aims to leverage the heat-reducing properties of the electric field to place as as many as hundreds of layers of memscapacitors on a single chip — greatly increasing compute capacity.

Semron

A schematic showing Semron’s 3D AI chip design.

“We use this property as an enabler to deploy several hundred times the compute resources on a fixed silicon area,” Kirschen added. “Think of it like hundreds of chips in one package.”

In a 2021 study published in the journal Nature Electronics, researchers at Semron and the Max Planck Institute of Microstructure Physics successfully trained a computer vision model at energy efficiencies of over 3,500 TOPS/W — 35 to 300 times higher than existing techniques. TOPS/W is a bit of a vague metric, but the takeaway is that memcapacitors can lead to dramatic energy consumption reductions while training AI models.

Now, it’s early days for Semron, which Kirschen says is in the “pre-product” stage and has “negligible” revenue to show for it. Often the toughest part of ramping up a chip startup is mass manufacturing and attaining a meaningful customer base — albeit not necessarily in that order.

Making matters more difficult for Semron is the fact that it has stiff competition in custom chip ventures like Kneron, EnCharge and Tenstorrent, which have collectively raised tens of millions of dollars in venture capital. EnCharge, like Semron, is designing computer chips that use capacitors rather than transistors, but using a different substrate architecture.

Semron, however — which has an 11-person workforce that it’s planning to grow by around 25 people by the end of the year — has managed to attract funding from investors including Join Capital, SquareOne, OTB Ventures and Onsight Ventures. To date, the startup has raised 10 million euro (~$10.81 million).

Said SquareOne partner Georg Stockinger via email:

“Computing resources will become the ‘oil’ of the 21st century. With infrastructure-hungry large language models conquering the world and Moore’s law reaching the limits of physics, a massive bottleneck in computing resources will shape the years to come. Insufficient access to computing infrastructure will greatly slow down productivity and competitiveness both of companies and entire nation-states. Semron will be a key element in solving this problem by providing a revolutionary new chip that is inherently specialized on computing AI models. It breaks with the traditional transistor-based computing paradigm and reduces costs and energy consumption for a given computing task by at least 20x.”



Source link

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Tech

ElevenLabs’ AI voice generation ‘very likely’ used in a Russian influence operation

Published

on

By


Generative AI has a plethora of well-documented misuses, from making up academic papers to copying artists. And now, it appears to be cropping up in state influence operations.

One recent campaign was “very likely” helped by commercial AI voice generation products, including tech publicly released by the hot startup ElevenLabs, according to a recent report from Massachusetts-based threat intelligence company Recorded Future.

The report describes a Russian-tied campaign designed to undermine Europe’s support for Ukraine, dubbed “Operation Undercut,” that prominently used AI-generated voiceovers on fake or misleading “news” videos.

The videos, which targeted European audiences, attacked Ukrainian politicians as corrupt or questioned the usefulness of military aid to Ukraine, among other themes. For example, one video touted that “even jammers can’t save American Abrams tanks,” referring to devices used by US tanks to deflect incoming missiles – reinforcing the point that sending high-tech armor to Ukraine is pointless.

The report states that the video creators “very likely” used voice-generated AI, including ElevenLabs tech, to make their content appear more legitimate. To verify this, Recorded Future’s researchers submitted the clips to ElevenLabs’ own AI Speech Classifier, which provides the ability for anyone to “detect whether an audio clip was created using ElevenLabs,” and got a match. 

ElevenLabs did not respond to requests for comment. Although Recorded Future noted the likely use of several commercial AI voice generation tools, it did not name any others besides ElevenLabs.

The usefulness of AI voice generation was inadvertently showcased by the influence campaign’s own orchestrators, who – rather sloppily – released some videos with real human voiceovers that had “a discernible Russian accent.” In contrast, the AI-generated voiceovers spoke in multiple European languages like English, French, German, and Polish, with no foreign-soundings accents.

According to Recorded Future, AI also allowed for the misleading clips to be quickly released in multiple languages spoken in Europe like English, German, French, Polish, and Turkish (incidentally, all languages supported by ElevenLabs.)

Recorded Future attributed the activity to the Social Design Agency, a Russia-based organization that the U.S. government sanctioned this March for running “ a network of over 60 websites that impersonated genuine news organizations in Europe, then used bogus social media accounts to amplify the misleading content of the spoofed websites.” All this was done “on behalf of the Government of the Russian Federation,” the U.S. State Department said at the time.

The overall impact of the campaign on public opinion in Europe was minimal, Recorded Future concluded.

This isn’t the first time ElevenLabs’ products have been singled out for alleged misuse.  The company’s tech was behind a robocall impersonating President Joe Biden that urged voters not to go out and vote during a primary election in January 2024, a voice fraud detection company concluded, according to Bloomberg. In response, ElevenLabs said it released new safety features like automatically blocking voices of politicians.

ElevenLabs bans “unauthorized, harmful, or deceptive impersonation” and says it uses various tools to enforce this, such as both automated and human moderation. 

ElevenLabs has experienced explosive growth since its founding in 2022. It recently grew ARR to $80 million from $25 million less than a year earlier, and may soon be valued at $3 billion, TechCrunch previously reported. Its investors include Andreessen Horowitz and former Github CEO Nat Friedman. 



Source link

Continue Reading

Tech

YouTube’s new auto-dubbing feature is now available for knowledge-focused content

Published

on

By


YouTube announced on Tuesday that its auto-dubbing feature, which allows creators to generate translated audio tracks for their videos, is now rolling out to hundreds of thousands more channels. 

YouTube first introduced its AI-powered auto-dubbing tool at Vidcon last year, which was only being tested with a limited group of creators. This tool could help make content on the platform more accessible and easier to understand for people all over the world.

The auto-dubbing feature is now available to channels that are focused on informational content, such as videos that teach viewers how to cook or sew. It’ll expand availability to other types of content soon. 

To use the feature, simply upload a video as you normally would. YouTube will automatically detect the language and create dubbed versions in other languages. The tool supports English, French, German, Hindi, Indonesian, Italian, Japanese, Portuguese, and Spanish.

Image Credits:YouTube

YouTube’s auto-dubbing utilizes Google’s Gemini capabilities to replicate human speech. However, the company cautions that the feature may not perform perfectly, as the technology is still in its early stages of development.

“We’re working hard to make it as accurate as possible, but there might be times when the translation isn’t quite right, or the dubbed voice doesn’t accurately represent the original speaker. We really appreciate your patience and feedback as we continue to improve,” the company wrote in Tuesday’s blog post. 

The company also reminded creators that they can look forward to another upcoming update called “Expressive Speech,” which is designed to help replicate the creator’s tone, emotions, and even the ambiance of their surroundings.



Source link

Continue Reading

Tech

CarDekho SEA raises first-ever outside funding, co-founder says he’s interested in acquisitions

Published

on

By


CarDekho SEA, a Singapore-based auto financing service platform, has raised $60 million in equity, valuing the company at over $300 million, from Navis Capital Partners and Dragon Fund. This is its first round of external funding, following a previous investment of $40 million from its parent company, CarDekho Group.

The company, the Southeast Asia unit of India’s CarDekho Group, will use the funding to support its further expansion into Southeast Asia, focusing on the used car and bike financing industry in Indonesia and the used auto financing sector in the Philippines, Umang Kumar, co-founder and president of CarDekho, said in an interview with TechCrunch.  

Kumar told TechCrunch that acquiring a used car financing platform or insurance brokerage platform in Indonesia and the Philippines could be one of its strategies to increase their presence in the region.

“We do have a couple of assets lined up in terms of what we will look at. You may hear something along those lines within a month or two…we will actively look at acquiring activities. The idea is not completely organically built. So we will look at some inorganic acquisition as we build out our business across Southeast,” Kumar said.

Founded in 2020, the company already has a significant presence in Indonesia (OTO Indonesia) and the Philippines (Carmudi Philippines and Zigwheels Philippines). It’s also in a few other Southeast Asian markets, such as Thailand, Singapore, Malaysia, and Vietnam, and is eyeing expansion into new markets in 2026.

Kumar explained that the company initially had a joint venture partner in Indonesia but later bought out the local partner entirely, becoming the 100% owner of the entity between 2019 and 2020. The company’s acquisition spree continues as it acquired Carmudi, which operates in the Philippines, Indonesia, and Thailand, in 2021 to expand into SEA, Kumar explained.

Unlike its peers, CarDekho SEO acts as a comprehensive aggregator for auto financing services. Its main competitor used to be Moladin, which has now pivoted to become a full-fledged lender.  

CarDekho SEA highlights its edges, including its technology-driven loan procedures, asset-light marketplace approach with no credit risk, an extensive network of used car dealers and agents, and solid collaborations with financial institutions. CarDekho SEA will use AI and machine learning solutions to reduce fraud and credit risks and provide more data to their financing partners to succeed in a changing technological environment.

Since its launch, the company has experienced significant growth, with over 200,000 disbursements and over $1 billion in loans disbursed. This represents a 50x increase in gross merchandise value over the past three years and establishing partnerships with over 50 financiers and 20,000 dealers and retail agents.

Its primary customers are small-scale used car dealers, with many having less than ten cars in stock or no physical showroom at all. “Due to their sub-scale presence, financiers often find it challenging to serve them directly,” the CEO said. “Our platform aggregates demand from these individual dealers, connecting them to a wide network of financiers and offering access to a variety of financial products for themselves and their customers—products that were previously inaccessible to them.”

The company is in the scale-up phase, generating upwards of $50 million in revenue annually through per-loan commissions with the financiers.



Source link

Continue Reading

Trending

Copyright © 2023 Dailycrunch. & Managed by Shade Marketing & PR Agency