Cadence Tools Paving the Way for the AI-at-the-Edge Transition

10 May 2023 • 7 minute read

For machines to sense and replicate human intelligence, we must empower them with quick decision-making abilities for reliable operations even during adverse conditions. AI at the edge reduces latency by processing data generated by sensors/IoT devices close to the source (edge), and there is no need to send the data to a centralized server/cloud. Low power, low bandwidth, and the least latency offerings of AI at the edge help make quick decisions. We must design and deliver chips with more computational power to process and analyze enormous amounts of data to enjoy the benefits. Design engineers and researchers are trying hard to make this possible! This blog post examines the industry trends of AI at the edge and how Cadence tools enable this transition.

Why AI Is Transitioning to the Edge?

We have seen drones offering delivery, robots for medicinal help, and autonomous vehicles for last-mile connectivity. AI at the edge promises to revolutionize our lives and is being adopted in many verticals such as autonomous driving, healthcare, manufacturing, retail industry, gaming, metaverse, AR/VR, financial/banking services, consumer applications, agriculture, defense, and drones. The Edge AI market fueled by AI-powered internet of things (IoT) advancements for intelligent systems and innovative applications is projected to reach US$8 billion by 2027. Innovations in semiconductor technology and design are driving this revolution.

It all started with using AI at a cloud/ data center for finding answers (inferencing) based on the training (data). However, the explosive data growth at the cloud/data centers due to increased IoT adoptions and mobile computing has resulted in issues such as latency and high bandwidth usage.

So, to unleash the full potential of big data and make accurate decisions in time, we need to process the data closer to the generation source. It can process diverse inputs like audio, video, and text with the least latency, and thus improve the intelligence and performance of applications. It offers multiple benefits, such as:

Real-time computing – Analyzing data locally rather than in a cloud enables the user to find answers (inferencing) in real time
Highly available due to decentralization and offline capability – Sensors and IoT applications generate data 24X7; the processing at the edge without data transfers to the cloud/servers
Robust and low cost – Edge AI is more robust, as internet access is not required for processing data, plus the low bandwidth and low data transfer require less expenditure
Privacy and security – Processing at the edge avoids the mishandling of data and enhances safety. Although it involves uploading some data for training purposes, user identities can be protected by anonymizing

Semiconductors and AI at the Edge

Semiconductors have become a new social infrastructure. AI at the edge started with mobile devices but can now be seen across all major applications, from healthcare to consumer products to autonomous driving. The ability of AI to drive performance is tied to semiconductor innovations. Semiconductors play a crucial role here, as AI and ML need powerful chips for smooth operation (inference and training). Further, many different workload facets demand semiconductor innovation to satisfy the needs of AI at the edge. To keep up, the industry is shifting from centralized SoCs to SoCs embedded with a specialized dedicated unit. These specialized SoCs, a.k.a. edge-based AI chips, consume less power and reduce latency. The advances in AI are leading to an exponential increase in semiconductor industry revenue; the edge-based AI chips driven by smartphones, metaverse (AR/VR), drones, robots, and industrial IoT (IIoT) are expected to reach $51.6 billion.

Source: Tractica

Challenges for AI at the Edge

Although moving AI to the edge offers many benefits, it comes with challenges, and with the increasing expectations and number of nodes, it is becoming complex. For instance, factors such as data sovereignty, control and interconnection, and data privacy are now more critical along with latency (initial perspective). Further, issues such as data preprocessing, storage, optimization, and meeting the varying computational and power requirements for different market needs make it hard to implement. Some of the challenges of moving AI over the edge include:

Lack of purpose-built silicon/hardware
Increasing complexity
Shrinking time windows
Maximizing performance/watt
Scalable solution designs
Safety and security for applications involved in life-critical decisions

How is the Industry Handling the Transition of AI at the Edge?

The industry is seeing a rise in applications involving automation. The insatiable hunger for improving business processes demands custom chips to process quickly and precisely. This is a moving target, and the industry is trying to keep up with the following solutions:

Using 3D packaging (heterogeneous architecture -chiplets)
New systems and memory architectures
Moving data centers close to the usage
Increasing the memory for handling the compute and processes
tinyML usage by the developers

How Cadence Supports the AI-at-the-Edge Transition

AI processing at the edge has become increasingly prevalent as it is adopted in various verticals and applications. Innovators have many on-device requirements for their AI-powered intellectual property (IP). Cadence’s computational software and the comprehensive on-device Tensilica platform help accelerate intelligent SoC development. In addition, Cadence EDA tools with AI capabilities enable faster verification than traditional verification schemes. These tools enhance the performance/watt for compute-intensive workloads and speed up the AI-at-the-edge transition. The Tensilica AI engine boosts performance, and AI accelerators provide a turnkey solution for consumer, mobile, automotive, and industrial AI SoC designs. Cadence solutions are widely used and successfully deployed in ubiquitous, high-volume AI-enabled end products, from low-cost voice-activated consumer devices to high-throughput autonomous vehicle perception. It helps customers deploy energy-efficient, intelligent, on-device edge computing solutions.

The Tensilica environment powers a full range of on-device AI solutions, including low-cost voice-activated consumer devices and high-throughput autonomous vehicle perception—based on the specific requirements for the intelligent sensor, IoT audio/vision, mobile, and automotive/ADAS markets.

The Cadence portfolio includes three platforms optimized for varying data and on-device AI requirements to provide optimal power, performance, and area (PPA) and a common software platform.

Cadence Tensilica Vision DSPs help quickly deploy AI systems on chip (SoCs) targeted for AI of things (AIoT), smart homes, intelligent surveillance, security, robotics, and industrial control applications. The "always-on" functionality requires clever AI algorithms that reliably detect events of interest. The Tensilica processor portfolio addresses the needs of ultra-low-power, always-on applications.

Our Integrity 3D-IC Platform is a “one-stop shop” solution that provides proven design flows for multi-chiplet design and advanced IC packaging. It provides 3D design planning, implementation, and system analysis in a unified cockpit. It enables hardware and software co-verification and full-system power analysis using emulation, prototyping, and chiplet-based PHY IP for connectivity with PPA optimized for latency, bandwidth, and power. The solution also offers co-design capabilities with custom analog design and board design, integrated circuit (IC) signoff extraction, static timing analysis (STA) and signoff with signal and power integrity (SI/PI), electromagnetic interference (EMI), and thermal analysis (more details in my previous post about 3D-IC trends).

Many of today's analog, RF, and mixed-signal designs require the integration of multiple ICs across varying substrate technologies to achieve the required performance goals. This heterogeneous integration introduces a new set of challenges for today's designers. Cadence also provides tools for low-power sensors and communication design at high frequency. The VirtuosoSystem Design Platform automates and streamlines multi-die heterogeneous systems' design and verification flow. Automating this flow eliminates the highly manual and error-prone process of integrating system-level layout parasitic models into the IC designer's flow.

Cadence's AWR Microwave Office Software helps develop high-frequency electronics. RF-aware layout, high-frequency models, and a powerful harmonic balance (HB) simulator ensure accurate and fast simulation results for first-pass success. The intuitive interface, innovative design automation, and design-assist features promote optimum engineering productivity and accelerate the development of RF/microwave IP across monolithic microwave IC (MMIC), RF PCB, and module technologies as standalone components or integrated within complex multi-technology systems.

Cadence offerings for AI hardware and software help answer the needs of various end applications, such as automotive, consumer, and smart homes. AI-embedded vision with depth perception and the 360-degree view help in accident prevention, decision-making, in-car assistance, etc. Cadence offerings at various levels are as follows:

Conclusion

We are witnessing the transition to an autonomous computing era fueled by AI over the edge. With the rising demands for quick decision-making and dependency on data, AI over the edge is here to stay. Rapid deployment is fundamental to take advantage of AI/ML in sectors such as defense, industry, healthcare, entertainment, and connectivity. Cadence provides comprehensive hardware, software platform, and IP to handle all workloads and markets.