DESPITE some unnecessary brouhaha about the Arm Holdings Inc deal with the Malaysian government, it is good to see more local companies making attempts to secure the licences.
Reports suggest that after the first round of awards in May, a second batch of local companies are close to receiving Arm’s technology under the programme.
Recall that the Malaysian government is buying access to Arm’s technology and know-how so that Malaysian companies can move from assembly and testing of chips to designing their own.
But the attempt to use the Arm IP and create something out of that is not an easy or cheap journey.
That long and costly journey to create commercially viable chips will take at least two years and cost around US$80mil to US$100mil.
So, before investors get over-exuberant about listed companies which are getting awarded the Arm tokens, they need to be aware of this challenge.
Interestingly, even Skyechip Bhd, in many ways the poster boy for Malaysia’s chip design ambitions, and which was among the first to be offered the Arm IP, is taking a cautious stand.
In its first-quarter result posted this week – its first quarterly result since its listing in May – Skyechip said it is still “reviewing and evaluating the terms and conditions of the definitive agreement for access to the compute subsystem (CSS) token”.
Skyechip had already inked the definitive agreement with the government for the Arm flexible access or AFA tokens.
To recap, the AFA entitles access to Arm’s vast IP library allowing them to experiment with and select the IP they need to design their own chips.
As for CSS, it provides a pre-designed and pre-assembled section of a chip that can serve as the foundation of a chip design, significantly reducing development time.
Using CSS tokens offer greater capabilities, but it also entails a costlier journey. This is why many of the recipients are looking to their customers to share some of those costs.
That, in turn, is not an easy feat.
The Arm tokens cover only around 20% to 30% of the overall chip development process.
Companies still have to bear significant costs for third-party IP, EDA software, computing infrastructure, manufacturing and engineering talent.
Despite the high costs, the Arm technology is crucial for Malaysia to move beyond its traditional back-end semiconductor activities and enter higher-value areas such as CPUs and AI chips, where the upfront barriers to entry are otherwise extremely high.
The government could potentially help commercialise these locally developed chips by extending existing data-centre local-content incentives to include locally developed semiconductor technologies.
This idea has been mooted before, but has yet to be formalised despite data centre investments continuing to rise.
The thing about the whole Arm programme is that while it is a heavy task, it is, in some ways, the only way ahead – that is, if Malaysian companies want to be part of the artificial intelligence revolution in any meaningful way.
What the Arm deal offers is a chance to move away from being merely a service provider to one that owns chip-related IP. Only then can a company occupy a healthy position in the tech value chain.
And while challenging, using the CSS tokens to create IP is not impossible – and this isn’t just a theoretical model. Take the case of Chinese automaker XPeng, which was reported earlier this year to be close to securing CSS token access alongside Skyechip, while Xenith Technology was said to be in a similar position with local startup GreatAsic Technology Sdn Bhd.
Automakers like these need
to develop their own Advanced Driver Assistance Systems, but building one from scratch is not their core competency. That is precisely the gap Malaysian chip designers are positioning themselves to fill: pairing Arm-based CSS access with a customer that already knows exactly what it needs built.
The ideal situation would be that this kind of customer also funds a big part of the developmental work and becomes the end-user of the chipsets created from the effort – which is essentially the arrangement Skyechip and XPeng appear to be moving towards.
Or, it could be a customer that wants its own customised CPU, where the local chip company then uses the Arm CSS to build out that project for the client.
The eureka moment will come when the local company is able to do this, as the profit margins will be phenomenal and the company will own all or at least a significant part of the IP that will be created.