Always a good sign when we get picked up by an Institution such as Macquarie. In my experience this is the first of what could be ongoing coverage. No recommendation yet. It will come though once they get a feel for the company and form a stronger bond.
Hi Belkin
Thank you for sharing this Macquarie analysis of Brainchip. We both share a long time holding tightly to our Brainchip investment and have had to suffer like many with the unsubstantiated attacks from HC trolls, the AFR, the MF, various analysts with short agendas, shorts and all round cheats, liars and fools.
Macquarie has sufficient gravitas in the Australian market that in my opinion the acceptance by it of certain matters as being fact should mean that no reasonable person will be lured into being influenced by those who in the past held sway claiming to be the voice of reason.
So as much for myself as for others I have extracted the following from the report by Macquarie, based on my own memory all the facts stated by Macquarie were constantly with almost a religious zeal disputed by the above mentioned cohort. The whole report should be read of course:
Key points:
BRN is a pure-play AI company, producing a neuromorphic processor called Akida.
Akida has utility across consumer and industrial applications, including autonomous vehicles, IoT devices, medical diagnosis.
Neuromorphic chip market is still nascent in a commercial sense, there is high competition & new technology risk from major players (e.g Intel & IBM)
Advantages of the Akida chip
Independence from cloud: Akida manages AI tasks at the Edge of the network instead of sending data to the cloud. Without needing an ongoing internet connection, Akida provides reduced system latency and faster response times
One shot learning: Reduces resources required to train models due to efficiency in accommodating new data inputs. Learns from very small set of samples and expands knowledge as more data is absorbed.
Low power consumption: Uses 100 microwatts to a few hundred milliwatts of power depending on the workload. It generates minimal heat from consuming low power,
so the chip should outlive the product it has been installed into.
On-chip fast learning and convolution: Traditional software-based neural networks (e.g CNN) can be efficiently run on BRN’s SNN by leveraging TensorFlow for industry standard neural network development.
Small and lightweight: 28nm chip can be designed into wearable devices and flying machines (e.g drones, aircraft, spacecraft).
Financials and risks
•
FY21 revenue of US$1.59M was a significant increase from US$120,829 in FY20, but still loss-making on EBITDA level.
•
Risks: IP, Reliance on key personnel, competition and new technologies, future funding requirements
Employs a Scientific Advisory Board (SAB) consisting of
three cognitive scientists and industry experts, including Nobel Prize Laureate Professor Barry J. Marshall, who joined in July 2020.
Directors’ background Management background
• Sean Hehir (CEO & Executive Director): Joined in November 2021,
a seasoned technology executive and board member of Silicon Valley Executive Network. Responsible for driving explosive revenue growth for HP, Compaq and Fusion-io.
• Peter van der Made (CTO & Executive Director): Co-founder of BrainChip. Previously CTO, founder and Chief Scientist of vCIS Technology, later Chief Scientist when acquired by Internet Security Systems and IBM. Also founded PolyGraphics Systems.
• Antonio Viana: On the board of Arteris, a leading provider of NoC interconnect. Previously Executive Chairman at QuantalRF,
former ARM President and EVP of Commercial and Global Development.
•
Geoffrey Carrick: On the board of Global Study Partners and VCF Capital Partners. Previously Head of Equity Capital Markets at CBA, Director of Equity Capital Markets at Macquarie Group and Head of Finance of Shaw & Partners.
• Pia Turcinov: Manages a portfolio career with qualifications in business management and law.
•
Kenneth W. Scarince (CFO): Joined in 2019. Previously held senior management positions as Finance Director of Midwest Connect, Controller of Virgin Galactic, VP and Controller of Virgin America, VP Finance of Chicago Express Airlines.
•
Anil Shamrao Mankar (CDO): Co-founder of BrainChip, previously CDO for Conexant Systems LLC, VP Business Development at T2M and Senior VP-VLSI Engineering at Mindspeed Technologies.
•
Jerome Nadel (CMO): Joined in January 2022 from Rambus, a NASDAQ-listed semiconductor technology company.
Previously held senior positions at Sagem, Thales, wireless and IoT solutions provider Option NV, Unisys and IBM. Board member of Silicon Valley Executive Network and President of Silicon Valley chapter of CMO club, a global community of marketing executives
Competition
BRN states it has a competitive advantage over rivals in the commercialisation process as other neuromorphic chips are still in the R&D stage.
IBM TrueNorth • Produced in 2014, a single processor consists of 5.4 billion transistors, 1 million neurons, 256 million synapses using 4,096 cores. Although it only uses milliwatts of power, each synapse needs to be programmed, which restricts the chip’s learning capabilities in real-time. Does not have backward compatibility with previous technology (e.g C++ compilers) and has vendor lock-in risks. ML workflow requires learning Corelet.
Intel Loihi • Introduced Loihi in January 2018 and its successor Loihi 2 in 2021, which provides 10x faster processing, 15x resource density and improved energy efficiency with a 7nm chip. Loihi 2 consists of 1 million neurons and 120 million synapses but lacks on-chip convolution and requires learning NEF for ML workflow. In 2019, Intel said they’re 5 years away from commercialisation.
Google Coral TPU • Introduced in 2019, Google Coral TPU is a ML application-specific integrated circuit (ASIC) designed to run AI at the edge. Provides high performance ML inferencing for low-power devices but only supports TensorFlow Lite. An individual Edge TPU performs 4 trillion (fixed-point) operations per second and consumes 2-5W of power.
DLAs (e.g Nvidia) • Launched in 2017, Nvidia produced an open-source hardware neural network AI accelerator written in Verilog. It is configurable and scalable to meet a range of architecture needs. However, as an accelerator, any process must be scheduled and arbitered by an outside entity (e.g CPU). Available for product development as a part of Nvidia’s Jetsen Xavier NX.
•
Crossed threshold from R&D into commercialisation: BRN evolved its operations to become the world’s first and only commercial producer of neuromorphic AI chips in FY21.
• Strong partnerships: BRN has IP licensing agreements with
Japan-based ASIC leader MegaChips and
global semiconductor manufacturer Renesas.
My opinion only DYOR
FF
AKIDA BALLISTA