Innovation has to return to the semiconductor industry at the transistor level rather than the exploitation of the massive transistor counts made possible by modern scaling, according to Andy Rappaport General Partner at August Capital.
Rappaport, who has invested in Actel, Atheros Genoa, MMC Networks, Silicon Architects (acquired by Synopsys) Silicon Image, Viewlogic and Transmeta has not invested in any start-up that proposed using 90nm or 65nm processes.
"The share of companies that will choose to develop very large chips will have to fall, if bulk transistor utilisation is expensive relative to value, marginal transistor improvement can be hugely valuable relative to cost," says Rappaport, who founded The Technology Research Group (TRG), and used to be an editor at EDN, a sister magazine to Electronics Weekly.
To put it another way, Rappaport says: "Lack of easy scaling makes clever scaling really important."
Rappaport points out that fabless semiconductor plays look easy at a time when tools are good, IP is plentiful, and dis-integration of customers increases the number of available sockets and success stories multiply, but looks hard when the cost of product realisation is 'not trivial', innovation is hard at the modern integration level, and when 'value is shifting back to increasing per-transistor work, not exploiting sheer transistor abundance'.
The statistics backing up Rappaport's thesis are interesting: The cost of fabless start-ups has risen from $10 million in the 1980s to over $30 million
VC investment in semiconductors peaked at $1.8 billion in 2001, slumped to $220 million in 2003 and has been running at an annual $400 million to $600 million a year ever since, according to Rappaport.
Exits during 2005-7 were valued at $4 billion, with marginally more coming from IPOs than M&A. Ten companies were acquired for less than $100 million, five for between $100 million and $249 million and three for $250 million to $300 million.
In the current decade there have been six IPOs at under $100 million, seven IPOs at between $100 million and $249 million, six IPOs at $250 million to $499 million, three IPOs at $500 million to $1 billion and two IPOs at over $1 billion.
This is not a good record, and something needs to be done to get the VCs interested in semiconductors again.
Rappaport concludes that the most valuable semiconductor start-ups will change the effective process curves by process augmentation, and by devising horizontally applicable circuits and architectures.
Leave a comment