AMD’s new AI accelerators; edge AI updates; microLED advancements; funding to private semi firms

Artificial intelligence takes center stage this week, with announcements from AMD, some edge AI news and the introduction of Gemini. Among other updates, the recent funding increase to private semiconductor firms.

AMD MI300 to challenge Nvidia H100

At the recent “Advancing AI” event, AMD has launched multiple new AI products, including the Instinct MI300 Series data center AI accelerators, ROCm 6 open software stack, and Ryzen 8040 Series processors with Ryzen AI. The event was intended to showcase growing momentum for AMD-powered AI solutions, particularly in data centers; on the occasion Microsoft, Dell Technologies, HPE, Lenovo, Meta, Oracle, Supermicro and others announced their adoption of the new AMD Instinct MI300X and MI300A data center AI accelerators for training and inference solutions. Analysis website SemiAnalysis has provided a benchmark-by-benchmark comparison of AMD MI300 performance versus Nvidia H100 GPU, highlighting that some AMD results only apply to the “forward pass” processing step. However, in terms of performance – including the software stack, a key element to take advantage of raw hardware power – SemiAnalysis sees AMD “rapidly improving.” SemiAnalysis also highlighted that OpenAI plans to support AMD’s GPUs, including MI300, in the standard Triton distribution starting with the upcoming 3.0 release.

AMD joins the “AI PC” race

Besides Intel, AMD is also touting the “AI PC” concept – that is, adding AI capabilities to mainstream x86 PC processors. The new AMD Ryzen 8040 Series mobile processors feature an integrated Ryzen AI NPU on-die on select models, offering up to 1.6x more AI processing performance than prior AMD models. AMD is also making Ryzen AI3 Software available for users to build and deploy machine learning models on their AI PCs. Laptops with Ryzen AI can offload AI models to the NPU, thereby freeing up the CPU to reduce power consumption while extending battery life. AMD Ryzen 8040 Series processors are expected to be available from OEMs including Acer, Asus, Dell, HP, Lenovo, and Razer, beginning in Q1 2024.

Edge AI updates: STMicroelectronics, Lattice, Infineon, Ceva

STMicroelectronics has announced the ST Edge AI Suite, an integrated set of software tools free-to-use with ST hardware. A key component of the ST Edge AI suite is the already existing ST Edge AI Core, which enables users to import ML and NN algorithms from the most widely used ML frameworks, analyze them, optimize the algorithm for the selected devices (sensors, MCU, MPU), validate against the original model, and map the resulting embedded AI solution on the selected device. Additionally, NanoEdge AI Studio autoML tool becomes free for STM32, and is now available for all Arm Cortex-M based MCUs The suite will be available in different fashions (desktop, CLI, web, API) from the first half of 2024. ST Edge AI Suite works across multiple ST hardware platforms, including STM32 MCUs, STM32N6 and STM32 MPUs, Stellar automotive microcontrollers, and smart sensors.

Lattice has introduced a new reference sensor-bridging design to accelerate the development of edge AI applications using the Nvidia Jetson Orin and IGX Orin platforms. The open-source reference board addresses connectivity to a variety of sensors and interfaces. The solution targets low latency, high-performance edge AI applications for healthcare, robotics, and embedded vision.

Imagimob, an Infineon company, is launching a series of machine learning “Ready Models” for edge devices. Ready Models can be deployed onto existing microcontroller hardware, such as PSoC 6. Imagimob is currently introducing four audio-based Ready Models: Baby Cry for baby monitors, Siren Detection for pedestrians, Coughing Detection and Snoring Detection for wearable devices. Additional models are under development for audio, radar, IMU (Inertial Measurement Unit) and capacitive sensing applications.

Ceva has introduced a new logo design and web domain (ceva-ip.com), reflecting the company’s commitment to IP solutions that power the smart edge. As the company stated in a press release, Ceva continues to sharpen its focus on delivering an innovative IP portfolio targeted at highly integrated, cost-efficient, low power Edge AI devices combining wireless connectivity, artificial intelligence and sensing technologies.

A 10km fiber, no DSP

In what the two companies describe as a milestone achievement, Intel new 224Gbps electrical SerDes design has been integrated with NewPhotonics Photonics Engine, resulting in an end-to-end direct modulation electrical-to-optical link utilizing PAM4 modulation. According to the companies, this accomplishment enables single-mode fiber of over ten kilometers with exceptional pre-FEC performance, without a need for a DSP, eliminating DSP delay and cost.

Color-tunable microLEDs

US startup Q-Pixel has developed what it claims is the world’s smallest full-color pixel and demonstrated it in the first ever 10,000 PPI, full-color microLED display. According to the company, 10,000 PPI resolution has previously been exhibited only in monochromatic, single-color displays. The achievement is based on Q-Pixel’s proprietary tunable polychromatic light emitting diode (TP-LED), a technology that enables the company to fabricate color tunable single pixels with diameters as small as 1 micron. This solution is meant to eliminate the costly and labor-intensive process of assembling full color microLED displays using monochromatic red, green, and blue LED subpixels.

According to market research firm Yole, microLED’s commercialization has faced delays, and significant production volumes are expected in 2-3 years with broader adoption in 5-10 years. As OLED technology advances, the urgency for microLED to deliver clear performance, functionality, and cost advantages has grown, says Yole.

Private semiconductor firms’ funding growth in Q3 2023

According to market intelligence firm CB Insights, equity funding to private semiconductor companies increased 68% quarter-over-quarter to $5.2B in Q3’23, marking the highest quarterly funding level since Q4’21, when it reached $7.8B. This is far beyond the growth seen in global venture funding, which increased by 11% QoQ in Q3’23. Meanwhile, deals continued a steady decline from their peak of 360 in Q3’21 to 188 in Q3’23.

Credit: CB Insights

In particular, the percentage of deals going to early-stage semiconductor startups has fallen to a multi-year low, says CB Insights.

Credit: CB Insights

CB Insights also released a ranking of the largest publicly-quoted semiconductor companies by market capitalization, which sees Nvidia at the first place – with TSMC a distant second.

Credit: CB Insights

Further reading

A detailed description of Gemini – Google’s new family of AI multimodal large language models, competing against OpenAI’s GPT-4 – can be found in this 62-page report.

This entry was posted on Monday, December 11th, 2023 at 2:09 am. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.