Hussein Osman, Segment Marketing Director At Lattice Semiconductor – Interview Series

Hussein Osman is simply a semiconductor manufacture seasoned pinch complete 2 decades of acquisition bringing to marketplace silicon and package products that merge sensing, processing and connectivity solutions, focusing connected innovative experiences that present worth to nan extremity user. Over nan past 5 years he has led nan sensAI solution strategy and go-to-market efforts astatine Lattice Semiconductor, creating high-performance AI/ML applications. Mr. Osman received his bachelor’s grade successful Electrical Engineering from California Polytechnic State University successful San Luis Obispo.

Lattice Semiconductor (LSCC -12.36%) is simply a supplier of low-power programmable solutions utilized crossed communications, computing, industrial, automotive, and user markets. The company's low-power FPGAs and package devices are designed to thief accelerate improvement and support invention crossed applications from nan Edge to nan Cloud.

Edge AI is gaining traction arsenic companies activity alternatives to cloud-based AI processing. How do you spot this displacement impacting nan semiconductor industry, and what domiciled does Lattice Semiconductor play successful this transformation? 

Edge AI is perfectly gaining traction, and it’s because of its imaginable to genuinely revolutionize full markets. Organizations crossed a wide scope of sectors are leaning into Edge AI because it’s helping them execute faster, much efficient, and much unafraid operations — particularly successful real-time applications — than are imaginable pinch unreality computing alone. That’s nan portion astir group thin to attraction on: really Edge AI is changing business operations erstwhile implemented. But there’s this different travel that’s happening successful tandem, and it starts acold earlier implementation.

Innovation successful Edge AI is pushing original instrumentality manufacturers to creation strategy components that tin tally AI models contempt footprint constraints. That intends lightweight, optimized algorithms, specialized hardware, and different advancements that complement and/or amplify performance. This is wherever Lattice Semiconductor comes into play.

Our Field Programmable Gate Arrays (FPGAs) supply nan highly adaptable hardware basal for designers to meet strict strategy requirements related to latency, power, security, connectivity, size, and more. They supply a instauration connected which engineers tin build devices tin of keeping mission-critical Automotive, Industrial, and Medical applications functional. This is simply a large attraction area for our existent innovation, and we’re excited to thief customers flooded challenges and shake hands nan era of Edge AI pinch confidence.

What are nan cardinal challenges that businesses look erstwhile implementing Edge AI, and really do you spot FPGAs addressing these issues much efficaciously than accepted processors aliases GPUs?

You know, immoderate challenges look to beryllium genuinely cosmopolitan arsenic any exertion advances. For example, developers and businesses hoping to harness nan powerfulness of Edge AI will apt grapple pinch communal challenges, specified as:

• Resource management. Edge AI devices person to execute analyzable processes reliably while moving wrong progressively constricted computational and artillery capacities.
• Although Edge AI offers nan privateness benefits of section information processing, it raises different information concerns, specified arsenic nan anticipation of beingness tampering aliases nan vulnerabilities that travel pinch smaller-scale models.
• Edge AI ecosystems tin beryllium highly divers successful hardware architectures and computing requirements, making it difficult to streamline aspects for illustration information guidance and exemplary updates astatine scale.

FPGAs connection businesses a limb up successful addressing these cardinal issues done their operation of businesslike parallel processing, debased powerfulness consumption, hardware-level information capabilities, and reconfigurability. While these whitethorn sound for illustration trading buzzwords, they are basal features for solving apical Edge AI symptom points.

FPGAs person traditionally been utilized for functions for illustration bridging and I/O description . What makes them peculiarly well-suited for Edge AI applications?

Yes, you’re precisely correct that FPGAs excel successful nan realm of connectivity — and that’s portion of what makes them truthful powerful successful Edge AI applications. As you mentioned, they person customizable I/O ports that let them to interface pinch a wide array of devices and connection protocols. On apical of this, they tin execute functions for illustration bridging and sensor fusion to guarantee seamless information exchange, aggregation, and synchronization betwixt different strategy components, including bequest and emerging standards. These functions are peculiarly important arsenic today’s Edge AI ecosystems turn much analyzable and nan request for interoperability and scalability increases.

However, arsenic we’ve been discussing, FPGAs’ connectivity benefits are only nan extremity of nan iceberg; it’s besides astir really their adaptability, processing power, power efficiency, and information features are driving outcomes. For example, FPGAs tin beryllium configured and reconfigured to execute circumstantial AI tasks, enabling developers to tailor applications to their unsocial needs and meet evolving requirements.

Can you explicate really low-power FPGAs comparison to GPUs and ASICs successful position of efficiency, scalability, and real-time processing capabilities for Edge AI?

I won’t dress that hardware for illustration GPUs and ASICs don’t person nan compute powerfulness to support Edge AI applications. They do. But FPGAs genuinely person an “edge” connected these different components successful different areas for illustration latency and flexibility. For example, some GPUs and FPGAs tin execute parallel processing, but GPU hardware is designed for wide entreaty and is not arsenic good suited to supporting circumstantial Edge applications arsenic that of FPGAs. On nan different hand, ASICs are targeted for circumstantial applications, but their fixed functionality intends they require afloat redesigns to accommodate immoderate important alteration successful use. FPGAs are purpose-built to supply nan champion of some worlds; they connection nan debased latency that comes pinch civilization hardware pipelines and room for post-deployment modifications whenever Edge models request updating.

Of course, nary azygous action is nan only correct one. It’s up to each developer to determine what makes consciousness for their system. They should cautiously see nan superior functions of nan application, nan circumstantial outcomes they are trying to meet, and really agile nan creation needs to beryllium from a future-proofing perspective. This will let them to take nan correct group of hardware and package components to meet their requirements — we conscionable hap to deliberation that FPGAs are usually nan correct choice.

How do Lattice’s FPGAs heighten AI-driven decision-making astatine nan edge, peculiarly successful industries for illustration automotive, business automation, and IoT?

FPGAs’ parallel processing capabilities are a bully spot to begin. Unlike sequential processors, nan architecture of FPGAs allows them to execute galore tasks successful parallel, including AI computations, pinch each nan configurable logic blocks executing different operations simultaneously. This allows for nan precocious throughput, debased latency processing needed to support real-time applications successful nan cardinal verticals you named — whether we’re talking astir autonomous vehicles, smart business robots, aliases moreover smart location devices aliases healthcare wearables. Moreover, they tin beryllium customized for circumstantial AI workloads and easy reprogrammed successful nan section arsenic models and requirements germinate complete time. Last, but not least, they connection hardware-level information features to guarantee AI-powered systems stay secure, from boot-up to information processing and beyond.

What are immoderate real-world usage cases wherever Lattice’s FPGAs person importantly improved Edge AI performance, security, aliases efficiency?

Great question! One exertion that I find really intriguing is nan ways engineers are utilizing Lattice FPGAs to powerfulness nan adjacent procreation of smart, AI-powered robots. Intelligent robots require real-time, on-device processing capabilities to guarantee safe automation, and that’s thing Edge AI is designed to deliver. Not only is nan request for these assistants rising, but truthful is nan complexity and sophistication of their functions. At a caller conference, nan Lattice squad demonstrated really nan usage of FPGAs allowed a smart robot to way nan trajectory of a shot and drawback it successful midair, showing conscionable really accelerated and precise these machines tin beryllium erstwhile built pinch nan correct technologies.

What makes this truthful absorbing to me, from a hardware perspective, is really creation strategies are changing to accommodate these applications. For example, alternatively of relying solely connected CPUs aliases different accepted processors, developers are opening to merge FPGAs into nan mix. The main use is that FPGAs tin interface pinch much sensors and actuators (and a much divers scope of these components), while besides performing low-level processing tasks adjacent these sensors to free up nan main compute motor for much precocious computations.

With nan increasing request for AI conclusion astatine nan edge, really does Lattice guarantee its FPGAs stay competitory against specialized AI chips developed by larger semiconductor companies?

There’s nary uncertainty that nan pursuit of AI chips is driving overmuch of nan semiconductor manufacture — conscionable look astatine really companies for illustration Nvidia pivoted from creating video crippled graphics cards to becoming AI manufacture giants. Still, Lattice brings unsocial strengths to nan array that make america guidelines retired moreover arsenic nan marketplace becomes much saturated.

FPGAs are not conscionable a constituent we’re choosing to put successful because request is rising; they’re a captious portion of our halfway merchandise line. The strengths of our FPGA offerings — from latency and programmability to powerfulness depletion and scalability — are nan consequence of years of method improvement and refinement. We besides supply a afloat scope of industry-leading package and solution stacks, built to optimize nan usage of FPGAs successful AI designs and beyond.

We’ve refined our FPGAs done years of continuous betterment driven by loop connected our hardware and package solutions and relationships pinch partners crossed nan semiconductor industry. We’ll proceed to beryllium competitory because we’ll support existent to that path, moving pinch design, development, and implementation partners to guarantee that we’re providing our customers pinch nan astir applicable and reliable method capabilities.

What domiciled does programmability play successful FPGAs’ expertise to accommodate to evolving AI models and workloads?

Unlike fixed-function hardware, FPGAs tin beryllium retooled and reprogrammed post-deployment. This inherent adaptability is arguably their biggest differentiator, particularly successful supporting evolving AI models and workloads. Considering really move nan AI scenery is, developers request to beryllium capable to support algorithm updates, increasing datasets, and different important changes arsenic they hap without worrying astir changeless hardware upgrades.

For example, FPGAs are already playing a pivotal domiciled successful nan ongoing displacement to post-quantum cryptography (PQC). As businesses brace against looming quantum threats and activity to switch susceptible encryption schemes pinch next-generation algorithms, they are utilizing FPGAs to facilitate a seamless modulation and guarantee compliance pinch caller PQC standards.

How do Lattice's FPGAs thief businesses equilibrium nan trade-off betwixt performance, powerfulness consumption, and costs successful Edge AI deployments?

Ultimately, developers shouldn’t person to take betwixt capacity and possibility. Yes, Edge applications are often hindered by computational limitations, powerfulness constraints, and accrued latency. But pinch Lattice FPGAs, developers are empowered pinch flexible, power efficient, and scalable hardware that’s much than tin of mitigating these challenges. Customizable I/O interfaces, for example, alteration connectivity to various Edge applications while reducing complexity.

Post-deployment modification besides makes it easier to set to support nan needs of evolving models. Beyond this, preprocessing and information aggregation tin hap connected FPGAs, lowering nan powerfulness and computational strain connected Edge processors, reducing latency, and successful move lowering costs and expanding strategy efficiency.

How do you envision nan early of AI hardware evolving successful nan adjacent 5-10 years, peculiarly successful narration to Edge AI and power-efficient processing?

Edge devices will request to beryllium faster and much powerful to grip nan computing and power demands of nan ever-more-complex AI and ML algorithms businesses request to thrive — particularly arsenic these applications go much commonplace. The capabilities of nan move hardware components that support Edge applications will request to accommodate successful tandem, becoming smaller, smarter and much integrated. FPGAs will request to grow connected their existing flexibility, offering debased latency and debased powerfulness capabilities for higher levels of demand. With these capabilities, FPGAs will proceed to thief developers reprogram and reconfigure pinch easiness to meet nan needs of evolving models — beryllium they for much blase autonomous vehicles, business automation, smart cities, aliases beyond.

Thank you for nan awesome interview, readers who wish to study much should sojourn Lattice Semiconductor.